# Multimodal benchmarks > Source: Benchize (https://benchize.com) — https://benchize.com/benchmarks/category/multimodal. When citing or reusing this content, credit "Benchize" and link to https://benchize.com/benchmarks/category/multimodal. Explore Multimodal LLM benchmarks and AI model results. Compare source-reported scores with original metrics, evaluation settings and evidence on Benchize. Canonical URL: https://benchize.com/benchmarks/category/multimodal Last updated: 2026-10-21T18:05:10.505Z Reproduction and quotation must name Benchize, link to the canonical HTML URL, and retain individual source citations. Evaluate models working with images, audio, video, and mixed inputs. Tasks range from recognizing content to reasoning over charts, documents, and other visual or audiovisual evidence. Published Oct 21, 2026 ## [Rapidata SVG generation — Preference](https://benchize.com/benchmarks/rapidata%3Asvg%3Arlb_1MFkkEMFnCNyKl.md) Updated: 10/21/2026 Models tested: 37 Human pairwise evaluation of svg outputs, measuring preference. Scores are fitted using the Bradley–Terry preference model and expressed on the publisher’s Elo scale with 95%… Top models · rapidata Latest result per model · v1 - 1[Gemini 4 Argon](https://benchize.com/models/gemini-4-argon.md) - 2[Claude Sonnet 5.5](https://benchize.com/models/claude-sonnet-5-5.md) - 3[GPT-6.1 Sol](https://benchize.com/models/gpt-6-1-sol.md) View details → ## [Rapidata SVG generation — Alignment](https://benchize.com/benchmarks/rapidata%3Asvg%3Arlb_1MFkk9yZHssERh.md) Updated: 10/21/2026 Models tested: 37 Human pairwise evaluation of svg outputs, measuring alignment. Scores are fitted using the Bradley–Terry preference model and expressed on the publisher’s Elo scale with 95%… Top models · rapidata Latest result per model · v1 - 1[Gemini 4 Argon](https://benchize.com/models/gemini-4-argon.md) - 2[Claude Sonnet 5.5](https://benchize.com/models/claude-sonnet-5-5.md) - 3[Claude Opus 5.5](https://benchize.com/models/claude-opus-5-5.md) View details → ## [Rapidata SVG generation — Coherence](https://benchize.com/benchmarks/rapidata%3Asvg%3Arlb_1MFkkBzpuU8fTG.md) Updated: 10/21/2026 Models tested: 37 Human pairwise evaluation of svg outputs, measuring coherence. Scores are fitted using the Bradley–Terry preference model and expressed on the publisher’s Elo scale with 95%… Top models · rapidata Latest result per model · v1 - 1[GPT-6 Astra](https://benchize.com/models/gpt-6-astra.md) - 2[GPT-6.1 Sol](https://benchize.com/models/gpt-6-1-sol.md) - 3[Claude Fable 5.1](https://benchize.com/models/claude-fable-5-1.md) View details → ## [Rapidata SVG generation — Overall preference](https://benchize.com/benchmarks/rapidata%3Asvg%3Aoverall.md) Updated: 10/21/2026 Models tested: 37 Human pairwise evaluation of svg outputs, measuring overall preference. Scores are fitted using the Bradley–Terry preference model and expressed on the publisher’s Elo scale with… Top models · rapidata Latest result per model · v1 - 1[GPT-6.1 Sol](https://benchize.com/models/gpt-6-1-sol.md) - 2[Claude Sonnet 5.5](https://benchize.com/models/claude-sonnet-5-5.md) - 3[Claude Opus 5.5](https://benchize.com/models/claude-opus-5-5.md) View details → ## [Arena vision / overall (config: vision)](https://benchize.com/benchmarks/benchmark%3Avision%3Aoverall--config-vision.md) Updated: 10/21/2026 Models tested: 82 Human pairwise preferences on the complete prompt population in Arena’s vision evaluation. The original organizer publishes model ratings and uncertainty; these are preference… Top models · Arena Latest result per model · Version not reported - 1[Claude Fable 5](https://benchize.com/models/claude-fable-5-high.md)High - 2[Claude Opus 5.5](https://benchize.com/models/claude-opus-5-5-high.md)High - 3[Claude Opus 5](https://benchize.com/models/claude-opus-5-high.md)High View details → ## [Arena vision style control / overall](https://benchize.com/benchmarks/lmarena%3Avision_style_control%3Aoverall.md) Updated: 10/21/2026 Models tested: 82 Human pairwise preferences on the complete prompt population in Arena’s vision style control evaluation. The original organizer publishes model ratings and uncertainty, adjusting… Top models · Arena Latest result per model · Version not reported - 1[Claude Fable 5](https://benchize.com/models/claude-fable-5-high.md)High - 2[Qwen3.8 Max](https://benchize.com/models/qwen3-8-max.md) - 3[Claude Opus 4.6](https://benchize.com/models/claude-opus-4-6-high.md)High View details → ## [Humanity's Sixth Sense](https://benchize.com/benchmarks/scale%3Ahss.md) Updated: 10/21/2026 Models tested: 24 Humans perceive far more in a scene than what is explicitly depicted: a single glance captures past causes and future trajectories; a quick peek determines if a vehicle can fit… Top models · scale Latest result per model · Version not reported - 1[Gemini 4 Argon](https://benchize.com/models/gemini-4-argon-high.md)High - 2[GPT-6 Astra](https://benchize.com/models/gpt-6-astra-max.md)Max - 3[GPT-6.1 Sol](https://benchize.com/models/gpt-6-1-sol-max.md)Max View details → ## [Furniture Assembly Benchmark](https://benchize.com/benchmarks/epoch-ai%3Afurniture_assembly.md) Updated: 10/21/2026 Models tested: 21 Detecting assembly mistakes from photographs of three IKEA furniture builds and their official assembly manuals. Models judge whether the build is correct and identify and explain… Top models · epoch ai Latest result per model · Furniture Assembly - 1[Claude Opus 5.5](https://benchize.com/models/claude-opus-5-5-max.md)Max - 2[GPT-6 Astra](https://benchize.com/models/gpt-6-astra-max.md)Max - 3[GPT-6.1 Sol](https://benchize.com/models/gpt-6-1-sol-max.md)Max View details → ## [SWE-bench Multimodal / bpmn js](https://benchize.com/benchmarks/vals-ai%3Aswebench-multimodal%3Abpmn_js.md) Updated: 10/21/2026 Models tested: 14 Resolving GitHub issues in JavaScript front-end repositories where the report is a screenshot as much as a description Top models · Vals AI Latest result per model · 1 - 1[Claude Opus 5.5](https://benchize.com/models/claude-opus-5-5-max.md)Max - 2[Claude Fable 5.1](https://benchize.com/models/claude-fable-5-1-max.md)Max - 3[Claude Sonnet 5.5](https://benchize.com/models/claude-sonnet-5-5-max.md)Max View details → ## [SWE-bench Multimodal / carbon](https://benchize.com/benchmarks/vals-ai%3Aswebench-multimodal%3Acarbon.md) Updated: 10/21/2026 Models tested: 14 Resolving GitHub issues in JavaScript front-end repositories where the report is a screenshot as much as a description Top models · Vals AI Latest result per model · 1 - 1[Claude Opus 5.5](https://benchize.com/models/claude-opus-5-5-max.md)Max - 2[Claude Fable 5.1](https://benchize.com/models/claude-fable-5-1-max.md)Max - 3[Gemini 4 Argon](https://benchize.com/models/gemini-4-argon-high.md)High View details → ## [SWE-bench Multimodal / eslint](https://benchize.com/benchmarks/vals-ai%3Aswebench-multimodal%3Aeslint.md) Updated: 10/21/2026 Models tested: 14 Resolving GitHub issues in JavaScript front-end repositories where the report is a screenshot as much as a description Top models · Vals AI Latest result per model · 1 - 1[Claude Opus 5.5](https://benchize.com/models/claude-opus-5-5-max.md)Max - 2[Claude Sonnet 5.5](https://benchize.com/models/claude-sonnet-5-5-max.md)Max - 3[Claude Fable 5.1](https://benchize.com/models/claude-fable-5-1-max.md)Max View details → ## [SWE-bench Multimodal / fusion next](https://benchize.com/benchmarks/vals-ai%3Aswebench-multimodal%3Afusion_next.md) Updated: 10/21/2026 Models tested: 14 Resolving GitHub issues in JavaScript front-end repositories where the report is a screenshot as much as a description Top models · Vals AI Latest result per model · 1 - 1[Claude Opus 5.5](https://benchize.com/models/claude-opus-5-5-max.md)Max - 2[Gemini 4 Argon](https://benchize.com/models/gemini-4-argon-high.md)High - 3[Claude Sonnet 5.5](https://benchize.com/models/claude-sonnet-5-5-max.md)Max View details → ## [SWE-bench Multimodal / grommet](https://benchize.com/benchmarks/vals-ai%3Aswebench-multimodal%3Agrommet.md) Updated: 10/21/2026 Models tested: 14 Resolving GitHub issues in JavaScript front-end repositories where the report is a screenshot as much as a description Top models · Vals AI Latest result per model · 1 - 1[Claude Opus 5.5](https://benchize.com/models/claude-opus-5-5-max.md)Max - 2[Gemini 4 Argon](https://benchize.com/models/gemini-4-argon-high.md)High - 3[Muse Wonder](https://benchize.com/models/muse-wonder-max.md)Max View details → ## [SWE-bench Multimodal / highlight js](https://benchize.com/benchmarks/vals-ai%3Aswebench-multimodal%3Ahighlight_js.md) Updated: 10/21/2026 Models tested: 14 Resolving GitHub issues in JavaScript front-end repositories where the report is a screenshot as much as a description Top models · Vals AI Latest result per model · 1 - 1[Claude Opus 5.5](https://benchize.com/models/claude-opus-5-5-max.md)Max - 2[Claude Fable 5.1](https://benchize.com/models/claude-fable-5-1-max.md)Max - 3[Claude Sonnet 5.5](https://benchize.com/models/claude-sonnet-5-5-max.md)Max View details → ## [SWE-bench Multimodal / lighthouse](https://benchize.com/benchmarks/vals-ai%3Aswebench-multimodal%3Alighthouse.md) Updated: 10/21/2026 Models tested: 14 Resolving GitHub issues in JavaScript front-end repositories where the report is a screenshot as much as a description Top models · Vals AI Latest result per model · 1 - 1[Claude Opus 5.5](https://benchize.com/models/claude-opus-5-5-max.md)Max - 2[Claude Fable 5.1](https://benchize.com/models/claude-fable-5-1-max.md)Max - 3[Claude Sonnet 5.5](https://benchize.com/models/claude-sonnet-5-5-max.md)Max View details → ## [SWE-bench Multimodal / openlayers](https://benchize.com/benchmarks/vals-ai%3Aswebench-multimodal%3Aopenlayers.md) Updated: 10/21/2026 Models tested: 14 Resolving GitHub issues in JavaScript front-end repositories where the report is a screenshot as much as a description Top models · Vals AI Latest result per model · 1 - 1[Claude Opus 5.5](https://benchize.com/models/claude-opus-5-5-max.md)Max - 2[Claude Fable 5.1](https://benchize.com/models/claude-fable-5-1-max.md)Max - 3[GPT-6.1 Sol](https://benchize.com/models/gpt-6-1-sol-max.md)Max View details → ## [SWE-bench Multimodal / overall](https://benchize.com/benchmarks/vals-ai%3Aswebench-multimodal%3Aoverall.md) Updated: 10/21/2026 Models tested: 14 Resolving GitHub issues in JavaScript front-end repositories where the report is a screenshot as much as a description Top models · Vals AI Latest result per model · 1 - 1[Claude Opus 5.5](https://benchize.com/models/claude-opus-5-5-max.md)Max - 2[Claude Fable 5.1](https://benchize.com/models/claude-fable-5-1-max.md)Max - 3[Claude Sonnet 5.5](https://benchize.com/models/claude-sonnet-5-5-max.md)Max View details → ## [SWE-bench Multimodal / prettier](https://benchize.com/benchmarks/vals-ai%3Aswebench-multimodal%3Aprettier.md) Updated: 10/21/2026 Models tested: 14 Resolving GitHub issues in JavaScript front-end repositories where the report is a screenshot as much as a description Top models · Vals AI Latest result per model · 1 - 1[Claude Opus 5.5](https://benchize.com/models/claude-opus-5-5-max.md)Max - 2[Claude Fable 5.1](https://benchize.com/models/claude-fable-5-1-max.md)Max - 3[Claude Sonnet 5.5](https://benchize.com/models/claude-sonnet-5-5-max.md)Max View details → ## [SWE-bench Multimodal / prism](https://benchize.com/benchmarks/vals-ai%3Aswebench-multimodal%3Aprism.md) Updated: 10/21/2026 Models tested: 14 Resolving GitHub issues in JavaScript front-end repositories where the report is a screenshot as much as a description Top models · Vals AI Latest result per model · 1 - 1[Claude Opus 5.5](https://benchize.com/models/claude-opus-5-5-max.md)Max - 2[Claude Sonnet 5.5](https://benchize.com/models/claude-sonnet-5-5-max.md)Max - 3[Claude Fable 5.1](https://benchize.com/models/claude-fable-5-1-max.md)Max View details → ## [SWE-bench Multimodal / scratch gui](https://benchize.com/benchmarks/vals-ai%3Aswebench-multimodal%3Ascratch_gui.md) Updated: 10/21/2026 Models tested: 14 Resolving GitHub issues in JavaScript front-end repositories where the report is a screenshot as much as a description Top models · Vals AI Latest result per model · 1 - 1[Claude Sonnet 5.5](https://benchize.com/models/claude-sonnet-5-5-max.md)Max - 2[Gemini 4 Argon](https://benchize.com/models/gemini-4-argon-high.md)High - 3[Claude Opus 5.5](https://benchize.com/models/claude-opus-5-5-max.md)Max View details → ## [Rapidata Text-to-3D with LLM code — Preference](https://benchize.com/benchmarks/rapidata%3Atext-to-3d-code%3Arlb_1YVpXugYHlm63x.md) Updated: 10/21/2026 Models tested: 9 Human pairwise evaluation of llm-to-3d outputs, measuring preference. Scores are fitted using the Bradley–Terry preference model and expressed on the publisher’s Elo scale with… Top models · rapidata Latest result per model · Version not reported - 1[Gemini 4 Argon](https://benchize.com/models/gemini-4-argon.md) - 2[GPT-6 Astra](https://benchize.com/models/gpt-6-astra.md) - 3[GPT-6.1 Sol](https://benchize.com/models/gpt-6-1-sol.md) View details → ## [Rapidata Text-to-3D with LLM code — Alignment](https://benchize.com/benchmarks/rapidata%3Atext-to-3d-code%3Arlb_1YVpXug2nDeUlV.md) Updated: 10/21/2026 Models tested: 9 Human pairwise evaluation of llm-to-3d outputs, measuring alignment. Scores are fitted using the Bradley–Terry preference model and expressed on the publisher’s Elo scale with 95%… Top models · rapidata Latest result per model · Version not reported - 1[GPT-6 Astra](https://benchize.com/models/gpt-6-astra.md) - 2[Gemini 4 Argon](https://benchize.com/models/gemini-4-argon.md) - 3[GPT-6.1 Sol](https://benchize.com/models/gpt-6-1-sol.md) View details → ## [Rapidata Text-to-3D with LLM code — Coherence](https://benchize.com/benchmarks/rapidata%3Atext-to-3d-code%3Arlb_1YVpXufudyOLlc.md) Updated: 10/21/2026 Models tested: 9 Human pairwise evaluation of llm-to-3d outputs, measuring coherence. Scores are fitted using the Bradley–Terry preference model and expressed on the publisher’s Elo scale with 95%… Top models · rapidata Latest result per model · Version not reported - 1[GPT-6 Astra](https://benchize.com/models/gpt-6-astra.md) - 2[GPT-6.1 Sol](https://benchize.com/models/gpt-6-1-sol.md) - 3[Claude Opus 5.5](https://benchize.com/models/claude-opus-5-5.md) View details → ## [Rapidata Text-to-3D with LLM code — Creativity](https://benchize.com/benchmarks/rapidata%3Atext-to-3d-code%3Arlb_1YVqTDVUylKrUp.md) Updated: 10/21/2026 Models tested: 9 Human pairwise evaluation of llm-to-3d outputs, measuring creativity. Scores are fitted using the Bradley–Terry preference model and expressed on the publisher’s Elo scale with… Top models · rapidata Latest result per model · Version not reported - 1[Gemini 4 Argon](https://benchize.com/models/gemini-4-argon.md) - 2[GPT-6 Astra](https://benchize.com/models/gpt-6-astra.md) - 3[Gemini 3.8 Flash](https://benchize.com/models/gemini-3-8-flash.md) View details → ## [Rapidata Text-to-3D with LLM code — Overall preference](https://benchize.com/benchmarks/rapidata%3Atext-to-3d-code%3Aoverall.md) Updated: 10/21/2026 Models tested: 9 Human pairwise evaluation of llm-to-3d outputs, measuring overall preference. Scores are fitted using the Bradley–Terry preference model and expressed on the publisher’s Elo scale… Top models · rapidata Latest result per model · Version not reported - 1[Gemini 4 Argon](https://benchize.com/models/gemini-4-argon.md) - 2[GPT-6 Astra](https://benchize.com/models/gpt-6-astra.md) - 3[GPT-6.1 Sol](https://benchize.com/models/gpt-6-1-sol.md) View details → ## [MMMU-Pro](https://benchize.com/benchmarks/vals-ai%3Ammmu%3Aoverall.md) Updated: 10/21/2026 Models tested: 29 Multimodal understanding on MMMU-Pro without Python augmentation. Top models · Vals AI Latest result per model · 1 - 1[Claude Fable 5.1](https://benchize.com/models/claude-fable-5-1-max.md)Max - 2[Claude Opus 5](https://benchize.com/models/claude-opus-5-max.md)Max - 3[Claude Fable 5](https://benchize.com/models/claude-fable-5-max.md)Max View details → ## [DesignArena Image generation](https://benchize.com/benchmarks/designarena%3Aimage.md) Updated: 10/21/2026 Models tested: 23 Human pairwise preference evaluation of image generation. Voters compare model outputs for the same request; published Elo ratings and separate model configurations are retained… Top models · designarena Latest result per model · public - 1[GPT Image 2.5 Sunburst](https://benchize.com/models/gpt-image-2-5-sunburst.md) - 2[GPT Image 2.5 Flare](https://benchize.com/models/gpt-image-2-5-flare.md) - 3[GPT Image 2](https://benchize.com/models/gpt-image-2.md) View details → ## [Vals Multimodal Index / corp fin v2](https://benchize.com/benchmarks/vals-ai%3Avals_multimodal_index%3Acorp_fin_v2.md) Updated: 10/21/2026 Models tested: 20 Benchmark consisting of a weighted performance across finance, coding, and education tasks. Showing the potential impact that LLM's can have on the economy. Top models · Vals AI Latest result per model · 1.2 - 1[Claude Opus 5](https://benchize.com/models/claude-opus-5.md) - 2[Muse Spark 1.1](https://benchize.com/models/muse-spark-1-1-xhigh.md)Extra high - 3[Muse Spark 1.2](https://benchize.com/models/muse-spark-1-2.md) View details → ## [Vals Multimodal Index / finance agent](https://benchize.com/benchmarks/vals-ai%3Avals_multimodal_index%3Afinance_agent.md) Updated: 10/21/2026 Models tested: 20 Benchmark consisting of a weighted performance across finance, coding, and education tasks. Showing the potential impact that LLM's can have on the economy. Top models · Vals AI Latest result per model · 1.2 - 1[Muse Spark 1.2](https://benchize.com/models/muse-spark-1-2.md) - 2[Gemini 3.5 Flash](https://benchize.com/models/gemini-3-5-flash-high.md)High - 3[Muse Spark 1.1](https://benchize.com/models/muse-spark-1-1-xhigh.md)Extra high View details → ## [Vals Multimodal Index / mortgage tax](https://benchize.com/benchmarks/vals-ai%3Avals_multimodal_index%3Amortgage_tax.md) Updated: 10/21/2026 Models tested: 20 Benchmark consisting of a weighted performance across finance, coding, and education tasks. Showing the potential impact that LLM's can have on the economy. Top models · Vals AI Latest result per model · 1.2 - 1[Claude Opus 5](https://benchize.com/models/claude-opus-5.md) - 2[Claude Opus 4.7](https://benchize.com/models/claude-opus-4-7-max.md)Max - 3[Claude Sonnet 5](https://benchize.com/models/claude-sonnet-5-max.md)Max View details → ## [Vals Multimodal Index / overall](https://benchize.com/benchmarks/vals-ai%3Avals_multimodal_index%3Aoverall.md) Updated: 10/21/2026 Models tested: 20 Benchmark consisting of a weighted performance across finance, coding, and education tasks. Showing the potential impact that LLM's can have on the economy. Top models · Vals AI Latest result per model · 1.2 - 1[Claude Fable 5](https://benchize.com/models/claude-fable-5-max.md)Max - 2[Claude Opus 5](https://benchize.com/models/claude-opus-5.md) - 3[GPT-5.6 Sol](https://benchize.com/models/gpt-5-6-sol-max.md)Max View details → ## [Vals Multimodal Index / sage](https://benchize.com/benchmarks/vals-ai%3Avals_multimodal_index%3Asage.md) Updated: 10/21/2026 Models tested: 20 Benchmark consisting of a weighted performance across finance, coding, and education tasks. Showing the potential impact that LLM's can have on the economy. Top models · Vals AI Latest result per model · 1.2 - 1[Claude Opus 4.7](https://benchize.com/models/claude-opus-4-7-max.md)Max - 2[Claude Opus 4.8](https://benchize.com/models/claude-opus-4-8-max.md)Max - 3[GPT-5.6 Sol](https://benchize.com/models/gpt-5-6-sol-max.md)Max View details → ## [Vals Multimodal Index / swebench](https://benchize.com/benchmarks/vals-ai%3Avals_multimodal_index%3Aswebench.md) Updated: 10/21/2026 Models tested: 20 Benchmark consisting of a weighted performance across finance, coding, and education tasks. Showing the potential impact that LLM's can have on the economy. Top models · Vals AI Latest result per model · 1.2 - 1[GPT-5.6 Sol](https://benchize.com/models/gpt-5-6-sol.md) - 2[Claude Fable 5](https://benchize.com/models/claude-fable-5.md) - 3[Claude Opus 5](https://benchize.com/models/claude-opus-5.md) View details → ## [Vals Multimodal Index / terminal bench 2 1](https://benchize.com/benchmarks/vals-ai%3Avals_multimodal_index%3Aterminal_bench_2_1.md) Updated: 10/21/2026 Models tested: 20 Benchmark consisting of a weighted performance across finance, coding, and education tasks. Showing the potential impact that LLM's can have on the economy. Top models · Vals AI Latest result per model · 1.2 - 1[GPT-5.6 Sol](https://benchize.com/models/gpt-5-6-sol-max.md)Max - 2[Claude Opus 5](https://benchize.com/models/claude-opus-5-high.md)High - 3[Claude Fable 5](https://benchize.com/models/claude-fable-5.md) View details → ## [Vals Multimodal Index / vibe code bench](https://benchize.com/benchmarks/vals-ai%3Avals_multimodal_index%3Avibe_code_bench.md) Updated: 10/21/2026 Models tested: 20 Benchmark consisting of a weighted performance across finance, coding, and education tasks. Showing the potential impact that LLM's can have on the economy. Top models · Vals AI Latest result per model · 1.2 - 1[Claude Fable 5](https://benchize.com/models/claude-fable-5-max.md)Max - 2[Claude Opus 4.8](https://benchize.com/models/claude-opus-4-8-max.md)Max - 3[GPT-5.6 Sol](https://benchize.com/models/gpt-5-6-sol-max.md)Max View details → ## [Rapidata General image — text coherence](https://benchize.com/benchmarks/rapidata%3Aimage%3Arlb_1RNU9vh78AqXX1.md) Updated: 10/21/2026 Models tested: 12 Human pairwise evaluation of text-to-image outputs, measuring text coherence. Scores are fitted using the Bradley–Terry preference model and expressed on the publisher’s Elo scale… Top models · rapidata Latest result per model · v1 - 1[Seedream 4.5](https://benchize.com/models/seedream-4-5.md) - 2[GPT Image 2.5 Sunburst](https://benchize.com/models/gpt-image-2-5-sunburst.md) - 3[GPT Image 1.5](https://benchize.com/models/gpt-image-1-5-high.md)High View details → ## [Rapidata General image — coherence](https://benchize.com/benchmarks/rapidata%3Aimage%3Arlb_1PetHDuThgAvpj.md) Updated: 10/21/2026 Models tested: 12 Human pairwise evaluation of text-to-image outputs, measuring coherence. Scores are fitted using the Bradley–Terry preference model and expressed on the publisher’s Elo scale with… Top models · rapidata Latest result per model · v1 - 1[GPT Image 2.5 Sunburst](https://benchize.com/models/gpt-image-2-5-sunburst.md) - 2[GPT Image 2.5 Flare](https://benchize.com/models/gpt-image-2-5-flare.md) - 3[Nano Banana 2.1](https://benchize.com/models/gemini-nano-banana-2-1.md) View details → ## [Rapidata General image — preference](https://benchize.com/benchmarks/rapidata%3Aimage%3Arlb_1Pf6xMzyixXVOB.md) Updated: 10/21/2026 Models tested: 12 Human pairwise evaluation of text-to-image outputs, measuring preference. Scores are fitted using the Bradley–Terry preference model and expressed on the publisher’s Elo scale… Top models · rapidata Latest result per model · v1 - 1[GPT Image 2.5 Sunburst](https://benchize.com/models/gpt-image-2-5-sunburst.md) - 2[GPT Image 1.5](https://benchize.com/models/gpt-image-1-5-high.md)High - 3[GPT Image 2.5 Flare](https://benchize.com/models/gpt-image-2-5-flare.md) View details → ## [Rapidata General image — alignment](https://benchize.com/benchmarks/rapidata%3Aimage%3Arlb_1PetHDQclk27pp.md) Updated: 10/21/2026 Models tested: 12 Human pairwise evaluation of text-to-image outputs, measuring alignment. Scores are fitted using the Bradley–Terry preference model and expressed on the publisher’s Elo scale with… Top models · rapidata Latest result per model · v1 - 1[GPT Image 2.5 Flare](https://benchize.com/models/gpt-image-2-5-flare.md) - 2[GPT Image 2.5 Sunburst](https://benchize.com/models/gpt-image-2-5-sunburst.md) - 3[GPT Image 1.5](https://benchize.com/models/gpt-image-1-5-high.md)High View details → ## [Rapidata General image — Overall preference](https://benchize.com/benchmarks/rapidata%3Aimage%3Aoverall.md) Updated: 10/21/2026 Models tested: 12 Human pairwise evaluation of text-to-image outputs, measuring overall preference. Scores are fitted using the Bradley–Terry preference model and expressed on the publisher’s Elo… Top models · rapidata Latest result per model · v1 - 1[GPT Image 2.5 Sunburst](https://benchize.com/models/gpt-image-2-5-sunburst.md) - 2[GPT Image 2.5 Flare](https://benchize.com/models/gpt-image-2-5-flare.md) - 3[GPT Image 1.5](https://benchize.com/models/gpt-image-1-5-high.md)High View details → ## [DesignArena Audio realism](https://benchize.com/benchmarks/designarena%3Aaudiorealismbench.md) Updated: 10/21/2026 Models tested: 4 Human pairwise preference evaluation of audio realism. Voters compare model outputs for the same request; published Elo ratings and separate model configurations are retained for… Top models · designarena Latest result per model · public - 1[Speech 2 8 Hd](https://benchize.com/models/speech-2-8-hd.md) - 2[Gemini 3.8 Flash TTS](https://benchize.com/models/gemini-3-8-flash-tts.md) - 3[S2.1 Pro](https://benchize.com/models/s2-1-pro.md) View details → ## [PerceptionBench](https://benchize.com/benchmarks/perceptionbench.md) Updated: 10/21/2026 Models tested: 19 Atomic visual perception across the ten capabilities defined by PerceptionBench. Top models · perceptionbench Latest result per model · ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e - 1[GPT-5.6 Sol](https://benchize.com/models/gpt-5-6-sol.md) - 2[Kimi K3](https://benchize.com/models/kimi-k3.md) - 3[Claude Fable 5](https://benchize.com/models/claude-fable-5.md) View details → ## [DesignArena Image editing](https://benchize.com/benchmarks/designarena%3Aimage-to-image.md) Updated: 10/21/2026 Models tested: 9 Human pairwise preference evaluation of image editing. Voters compare model outputs for the same request; published Elo ratings and separate model configurations are retained for… Top models · designarena Latest result per model · public - 1[GPT Image 2.5 Sunburst](https://benchize.com/models/gpt-image-2-5-sunburst.md) - 2[GPT Image 2.5 Flare](https://benchize.com/models/gpt-image-2-5-flare.md) - 3[GPT Image 2](https://benchize.com/models/gpt-image-2.md) View details → ## [AudioMultiChallenge - Text Output](https://benchize.com/benchmarks/scale%3Aaudiomc-text.md) Updated: 10/21/2026 Models tested: 8 Audio MultiChallenge is a multi-turn benchmark that evaluates conversational intelligence in spoken dialogue systems and audio-language models, including both speech-to-speech and… Top models · scale Latest result per model · Version not reported - 1[Gemini 3.8 Flash](https://benchize.com/models/gemini-3-8-flash-high.md)High - 2[Gemini 4 Argon](https://benchize.com/models/gemini-4-argon-high.md)High - 3[Gemini 2.5 Pro](https://benchize.com/models/gemini-2-5-pro-thinking.md)Thinking View details → ## [AudioMultiChallenge](https://benchize.com/benchmarks/scale%3Aaudiomc.md) Updated: 10/21/2026 Models tested: 6 Audio MultiChallenge is a multi-turn benchmark that evaluates conversational intelligence in spoken dialogue systems and audio-language models, including both speech-to-speech and… Top models · scale Latest result per model · Version not reported - 1[Gemini 3.8 Flash](https://benchize.com/models/gemini-3-8-flash-high.md)High - 2[Gemini 4 Argon](https://benchize.com/models/gemini-4-argon-high.md)High - 3[Voxtral Small 24B 2507](https://benchize.com/models/voxtral-small-24b-2507.md) View details → ## [BabyVision w/ python](https://benchize.com/benchmarks/babyvision-python.md) Updated: 10/21/2026 Models tested: 5 Fine-grained visual perception and reasoning with Python assistance. Top models · moonshot k3 Latest result per model · sha256:0e806bbad76dc8e8dbc899a86051e3406bdd5995fe6c6aa1b2176dde4fb28e6c - 1[Claude Fable 5](https://benchize.com/models/claude-fable-5-max.md)Max - 2[GPT-5.6 Sol](https://benchize.com/models/gpt-5-6-sol-max.md)Max - 3[Kimi K3](https://benchize.com/models/kimi-k3-max.md)Max View details → ## [CharXiv (RQ)](https://benchize.com/benchmarks/charxiv-rq.md) Updated: 10/21/2026 Models tested: 5 Reasoning questions about scientific charts in the CharXiv RQ scope without Python assistance. Top models · moonshot k3 Latest result per model · sha256:0e806bbad76dc8e8dbc899a86051e3406bdd5995fe6c6aa1b2176dde4fb28e6c - 1[Claude Fable 5](https://benchize.com/models/claude-fable-5-max.md)Max - 2[Kimi K3](https://benchize.com/models/kimi-k3-max.md)Max - 3[GPT-5.6 Sol](https://benchize.com/models/gpt-5-6-sol-max.md)Max View details → ## [CharXiv (RQ) w/ python](https://benchize.com/benchmarks/charxiv-rq-python.md) Updated: 10/21/2026 Models tested: 5 CharXiv RQ scientific-chart questions with Python assistance. Top models · moonshot k3 Latest result per model · sha256:0e806bbad76dc8e8dbc899a86051e3406bdd5995fe6c6aa1b2176dde4fb28e6c - 1[Claude Fable 5](https://benchize.com/models/claude-fable-5-max.md)Max - 2[Kimi K3](https://benchize.com/models/kimi-k3-max.md)Max - 3[Claude Opus 4.8](https://benchize.com/models/claude-opus-4-8-max.md)Max View details → ## [MathVision](https://benchize.com/benchmarks/mathvision.md) Updated: 10/21/2026 Models tested: 5 Solving visual mathematics problems without Python assistance. Top models · moonshot k3 Latest result per model · sha256:0e806bbad76dc8e8dbc899a86051e3406bdd5995fe6c6aa1b2176dde4fb28e6c - 1[GPT-5.6 Sol](https://benchize.com/models/gpt-5-6-sol-max.md)Max - 2[Claude Fable 5](https://benchize.com/models/claude-fable-5-max.md)Max - 3[Kimi K3](https://benchize.com/models/kimi-k3-max.md)Max View details → ## [MathVision w/ python](https://benchize.com/benchmarks/mathvision-python.md) Updated: 10/21/2026 Models tested: 5 Solving visual mathematics problems with Python assistance. Top models · moonshot k3 Latest result per model · sha256:0e806bbad76dc8e8dbc899a86051e3406bdd5995fe6c6aa1b2176dde4fb28e6c - 1[Claude Fable 5](https://benchize.com/models/claude-fable-5-max.md)Max - 2[GPT-5.6 Sol](https://benchize.com/models/gpt-5-6-sol-max.md)Max - 3[Kimi K3](https://benchize.com/models/kimi-k3-max.md)Max View details → ## [MMMU-Pro w/ python](https://benchize.com/benchmarks/mmmu-pro-python.md) Updated: 10/21/2026 Models tested: 5 MMMU-Pro with Python assistance, following the source’s disclosed image-order protocol. Top models · moonshot k3 Latest result per model · sha256:0e806bbad76dc8e8dbc899a86051e3406bdd5995fe6c6aa1b2176dde4fb28e6c - 1[Claude Fable 5](https://benchize.com/models/claude-fable-5-max.md)Max - 2[GPT-5.6 Sol](https://benchize.com/models/gpt-5-6-sol-max.md)Max - 3[Kimi K3](https://benchize.com/models/kimi-k3-max.md)Max View details → ## [OmniDocBench](https://benchize.com/benchmarks/omnidocbench.md) Updated: 10/21/2026 Models tested: 5 Understanding complex document layouts in the OmniDocBench evaluation. Top models · moonshot k3 Latest result per model · sha256:0e806bbad76dc8e8dbc899a86051e3406bdd5995fe6c6aa1b2176dde4fb28e6c - 1[Kimi K3](https://benchize.com/models/kimi-k3-max.md)Max - 2[Claude Fable 5](https://benchize.com/models/claude-fable-5-max.md)Max - 3[GPT-5.5](https://benchize.com/models/gpt-5-5-xhigh.md)Extra high View details → ## [WorldVQA ForceAnswer](https://benchize.com/benchmarks/worldvqa-force-answer.md) Updated: 10/21/2026 Models tested: 5 Visual world-knowledge questions evaluated under the source’s forced-answer protocol. Top models · moonshot k3 Latest result per model · sha256:0e806bbad76dc8e8dbc899a86051e3406bdd5995fe6c6aa1b2176dde4fb28e6c - 1[Claude Fable 5](https://benchize.com/models/claude-fable-5-max.md)Max - 2[Kimi K3](https://benchize.com/models/kimi-k3-max.md)Max - 3[GPT-5.6 Sol](https://benchize.com/models/gpt-5-6-sol-max.md)Max View details → ## [ZeroBench main (pass@5)](https://benchize.com/benchmarks/zero-bench-main.md) Updated: 10/21/2026 Models tested: 5 ZeroBench main-set pass@5 over five runs without Python assistance. Top models · moonshot k3 Latest result per model · sha256:0e806bbad76dc8e8dbc899a86051e3406bdd5995fe6c6aa1b2176dde4fb28e6c - 1[Kimi K3](https://benchize.com/models/kimi-k3-max.md)Max - 2[Claude Fable 5](https://benchize.com/models/claude-fable-5-max.md)Max - 3[GPT-5.5](https://benchize.com/models/gpt-5-5-xhigh.md)Extra high View details → ## [ZeroBench main w/ python (pass@5)](https://benchize.com/benchmarks/zero-bench-main-python.md) Updated: 10/21/2026 Models tested: 5 ZeroBench main-set pass@5 over five runs with Python assistance. Top models · moonshot k3 Latest result per model · sha256:0e806bbad76dc8e8dbc899a86051e3406bdd5995fe6c6aa1b2176dde4fb28e6c - 1[Claude Fable 5](https://benchize.com/models/claude-fable-5-max.md)Max - 2[GPT-5.5](https://benchize.com/models/gpt-5-5-xhigh.md)Extra high - 3[Kimi K3](https://benchize.com/models/kimi-k3-max.md)Max View details → ## [DesignArena Text to speech](https://benchize.com/benchmarks/designarena%3Atts.md) Updated: 10/21/2026 Models tested: 2 Human pairwise preference evaluation of text to speech. Voters compare model outputs for the same request; published Elo ratings and separate model configurations are retained for… Top models · designarena Latest result per model · public - 1[Gemini 3.8 Flash TTS](https://benchize.com/models/gemini-3-8-flash-tts.md) - 2[Voxtral Mini TTS](https://benchize.com/models/voxtral-mini-tts-2603.md) View details → ## [VisualToolBench (VTB)](https://benchize.com/benchmarks/scale%3Avtb.md) Updated: 10/21/2026 Models tested: 5 VisualToolBench (VTB)is the first benchmark designed to evaluate how well Multimodal Large Language Models (MLLMs) can dynamically interact with and reason about visual… Top models · scale Latest result per model · Version not reported - 1[Muse Spark 1.1](https://benchize.com/models/muse-spark-1-1.md) - 2[Gemini 3.1 Pro Preview](https://benchize.com/models/gemini-3-1-pro-preview.md) - 3[Gemini 2.5 Pro Preview 06-05](https://benchize.com/models/gemini-2-5-pro-preview.md) View details → ## [Rapidata World Model Physics — with-reference](https://benchize.com/benchmarks/rapidata%3Aworld-model-physics%3Arlb_1U6OnkM0D4HJRL.md) Updated: 10/21/2026 Models tested: 4 Human pairwise evaluation of physics outputs, measuring with-reference. Scores are fitted using the Bradley–Terry preference model and expressed on the publisher’s Elo scale with… Top models · rapidata Latest result per model · v1 - 1[Happyhorse 1 0](https://benchize.com/models/happyhorse-1-0.md) - 2[Veo 3.1 Fast](https://benchize.com/models/veo-3-1-fast.md) - 3[Wan 2.7](https://benchize.com/models/wan-2-7.md) View details → ## [Rapidata World Model Physics — looks-like-reference-pinned](https://benchize.com/benchmarks/rapidata%3Aworld-model-physics%3Arlb_1XwtptXwuxJVMY.md) Updated: 10/21/2026 Models tested: 4 Human pairwise evaluation of physics outputs, measuring looks-like-reference-pinned. Scores are fitted using the Bradley–Terry preference model and expressed on the publisher’s… Top models · rapidata Latest result per model · v1 - 1[Happyhorse 1 0](https://benchize.com/models/happyhorse-1-0.md) - 2[Wan 2.6](https://benchize.com/models/wan-2-6.md) - 3[Wan 2.7](https://benchize.com/models/wan-2-7.md) View details → ## [Rapidata World Model Physics — with-reference-pinned](https://benchize.com/benchmarks/rapidata%3Aworld-model-physics%3Arlb_1Xwr15RsywqJgE.md) Updated: 10/21/2026 Models tested: 4 Human pairwise evaluation of physics outputs, measuring with-reference-pinned. Scores are fitted using the Bradley–Terry preference model and expressed on the publisher’s Elo… Top models · rapidata Latest result per model · v1 - 1[Happyhorse 1 0](https://benchize.com/models/happyhorse-1-0.md) - 2[Wan 2.6](https://benchize.com/models/wan-2-6.md) - 3[Veo 3.1 Fast](https://benchize.com/models/veo-3-1-fast.md) View details → ## [Rapidata World Model Physics — blind](https://benchize.com/benchmarks/rapidata%3Aworld-model-physics%3Arlb_1U6Onl3Z5Mz8Jz.md) Updated: 10/21/2026 Models tested: 4 Human pairwise evaluation of physics outputs, measuring blind. Scores are fitted using the Bradley–Terry preference model and expressed on the publisher’s Elo scale with 95%… Top models · rapidata Latest result per model · v1 - 1[Happyhorse 1 0](https://benchize.com/models/happyhorse-1-0.md) - 2[Veo 3.1 Fast](https://benchize.com/models/veo-3-1-fast.md) - 3[Wan 2.7](https://benchize.com/models/wan-2-7.md) View details → ## [Rapidata World Model Physics — Overall preference](https://benchize.com/benchmarks/rapidata%3Aworld-model-physics%3Aoverall.md) Updated: 10/21/2026 Models tested: 4 Human pairwise evaluation of physics outputs, measuring overall preference. Scores are fitted using the Bradley–Terry preference model and expressed on the publisher’s Elo scale… Top models · rapidata Latest result per model · v1 - 1[Happyhorse 1 0](https://benchize.com/models/happyhorse-1-0.md) - 2[Wan 2.6](https://benchize.com/models/wan-2-6.md) - 3[Veo 3.1 Fast](https://benchize.com/models/veo-3-1-fast.md) View details → ## [VISTA](https://benchize.com/benchmarks/scale%3Avisual_language_understanding.md) Updated: 10/21/2026 Models tested: 16 The growing capabilities of multimodal language models have created two critical challenges in their evaluation. First, as these models become more prevalent, their potential… Top models · scale Latest result per model · Version not reported - 1[Gemini 2.5 Pro Preview 06-05](https://benchize.com/models/gemini-2-5-pro-preview.md) - 2[Claude Opus 4](https://benchize.com/models/claude-opus-4-thinking.md)Thinking - 3[Gemini 3.1 Flash Lite Preview](https://benchize.com/models/gemini-3-1-flash-lite-preview.md) View details → ## [DesignArena Video generation](https://benchize.com/benchmarks/designarena%3Avideo.md) Updated: 10/21/2026 Models tested: 7 Human pairwise preference evaluation of video generation. Voters compare model outputs for the same request; published Elo ratings and separate model configurations are retained… Top models · designarena Latest result per model · public - 1[Seedance 2 0](https://benchize.com/models/seedance-2-0.md) - 2[Seedance 2.0 Fast](https://benchize.com/models/seedance-2-0-fast.md) - 3[Veo 3 1 Lite](https://benchize.com/models/veo-3-1-lite.md) View details → ## [DesignArena Image to video](https://benchize.com/benchmarks/designarena%3Aimage-to-video.md) Updated: 10/21/2026 Models tested: 6 Human pairwise preference evaluation of image to video. Voters compare model outputs for the same request; published Elo ratings and separate model configurations are retained for… Top models · designarena Latest result per model · public - 1[Seedance 2 0](https://benchize.com/models/seedance-2-0.md) - 2[Seedance 2.0 Fast](https://benchize.com/models/seedance-2-0-fast.md) - 3[Wan 2.7](https://benchize.com/models/wan-2-7.md) View details → ## [DesignArena Multiple images to video](https://benchize.com/benchmarks/designarena%3Amulti-to-video.md) Updated: 10/21/2026 Models tested: 5 Human pairwise preference evaluation of multiple images to video. Voters compare model outputs for the same request; published Elo ratings and separate model configurations are… Top models · designarena Latest result per model · public - 1[Seedance 2 0](https://benchize.com/models/seedance-2-0.md) - 2[Seedance 2.0 Fast](https://benchize.com/models/seedance-2-0-fast.md) - 3[Veo 3.1 Fast](https://benchize.com/models/veo-3-1-fast.md) View details → ## [DesignArena Multimodal video generation](https://benchize.com/benchmarks/designarena%3Amultimodal-to-video.md) Updated: 10/21/2026 Models tested: 3 Human pairwise preference evaluation of multimodal video generation. Voters compare model outputs for the same request; published Elo ratings and separate model configurations are… Top models · designarena Latest result per model · public - 1[Seedance 2.0 Fast](https://benchize.com/models/seedance-2-0-fast.md) - 2[Seedance 2 0](https://benchize.com/models/seedance-2-0.md) - 3[Wan 2.7](https://benchize.com/models/wan-2-7.md) View details → ## [Arena vision / english (config: vision)](https://benchize.com/benchmarks/benchmark%3Avision%3Aenglish--config-vision.md) Updated: 10/21/2026 Models tested: 82 Human pairwise preferences on english prompts in Arena’s vision evaluation. The original organizer publishes model ratings and uncertainty; these are preference ratings rather… Top models · Arena Latest result per model · Version not reported - 1[Claude Opus 4.6](https://benchize.com/models/claude-opus-4-6-high.md)High - 2[Claude Fable 5](https://benchize.com/models/claude-fable-5-high.md)High - 3[Claude Opus 4.7](https://benchize.com/models/claude-opus-4-7.md) View details → ## [Arena vision style control / english](https://benchize.com/benchmarks/lmarena%3Avision_style_control%3Aenglish.md) Updated: 10/21/2026 Models tested: 82 Human pairwise preferences on english prompts in Arena’s vision style control evaluation. The original organizer publishes model ratings and uncertainty, adjusting for response… Top models · Arena Latest result per model · Version not reported - 1[Claude Fable 5](https://benchize.com/models/claude-fable-5-high.md)High - 2[Claude Opus 4.7](https://benchize.com/models/claude-opus-4-7-high.md)High - 3[Claude Opus 4.7](https://benchize.com/models/claude-opus-4-7.md) View details → ## [Arena vision / diagram (config: vision)](https://benchize.com/benchmarks/benchmark%3Avision%3Adiagram--config-vision.md) Updated: 10/21/2026 Models tested: 74 Human pairwise preferences on diagram prompts in Arena’s vision evaluation. The original organizer publishes model ratings and uncertainty; these are preference ratings rather… Top models · Arena Latest result per model · Version not reported - 1[Claude Fable 5](https://benchize.com/models/claude-fable-5-high.md)High - 2[Claude Opus 5](https://benchize.com/models/claude-opus-5-high.md)High - 3[Claude Opus 5.5](https://benchize.com/models/claude-opus-5-5-high.md)High View details → ## [Arena vision / ocr (config: vision)](https://benchize.com/benchmarks/benchmark%3Avision%3Aocr--config-vision.md) Updated: 10/21/2026 Models tested: 74 Human pairwise preferences on ocr prompts in Arena’s vision evaluation. The original organizer publishes model ratings and uncertainty; these are preference ratings rather than… Top models · Arena Latest result per model · Version not reported - 1[Claude Fable 5](https://benchize.com/models/claude-fable-5-high.md)High - 2[Gemini 3.8 Flash](https://benchize.com/models/gemini-3-8-flash-high.md)High - 3[Claude Opus 5](https://benchize.com/models/claude-opus-5-high.md)High View details → ## [Arena vision style control / diagram](https://benchize.com/benchmarks/lmarena%3Avision_style_control%3Adiagram.md) Updated: 10/21/2026 Models tested: 74 Human pairwise preferences on diagram prompts in Arena’s vision style control evaluation. The original organizer publishes model ratings and uncertainty, adjusting for response… Top models · Arena Latest result per model · Version not reported - 1[Claude Fable 5](https://benchize.com/models/claude-fable-5-high.md)High - 2[Gemini 3.8 Flash](https://benchize.com/models/gemini-3-8-flash-high.md)High - 3[Claude Opus 4.7](https://benchize.com/models/claude-opus-4-7-high.md)High View details → ## [Arena vision style control / ocr](https://benchize.com/benchmarks/lmarena%3Avision_style_control%3Aocr.md) Updated: 10/21/2026 Models tested: 74 Human pairwise preferences on ocr prompts in Arena’s vision style control evaluation. The original organizer publishes model ratings and uncertainty, adjusting for response style;… Top models · Arena Latest result per model · Version not reported - 1[Claude Fable 5](https://benchize.com/models/claude-fable-5-high.md)High - 2[Gemini 3.8 Flash](https://benchize.com/models/gemini-3-8-flash-high.md)High - 3[Qwen3.8 Max](https://benchize.com/models/qwen3-8-max.md) View details → ## [Arena vision / homework (config: vision)](https://benchize.com/benchmarks/benchmark%3Avision%3Ahomework--config-vision.md) Updated: 10/21/2026 Models tested: 73 Human pairwise preferences on homework prompts in Arena’s vision evaluation. The original organizer publishes model ratings and uncertainty; these are preference ratings rather… Top models · Arena Latest result per model · Version not reported - 1[Claude Opus 5.5](https://benchize.com/models/claude-opus-5-5-high.md)High - 2[Claude Fable 5](https://benchize.com/models/claude-fable-5-high.md)High - 3[GPT-6.1 Sol](https://benchize.com/models/gpt-6-1-sol-max.md)Max View details → ## [Arena vision style control / homework](https://benchize.com/benchmarks/lmarena%3Avision_style_control%3Ahomework.md) Updated: 10/21/2026 Models tested: 73 Human pairwise preferences on homework prompts in Arena’s vision style control evaluation. The original organizer publishes model ratings and uncertainty, adjusting for response… Top models · Arena Latest result per model · Version not reported - 1[GPT-6.1 Sol](https://benchize.com/models/gpt-6-1-sol-max.md)Max - 2[Claude Fable 5](https://benchize.com/models/claude-fable-5-high.md)High - 3[Claude Opus 5.5](https://benchize.com/models/claude-opus-5-5-high.md)High View details → ## [Arena vision / creative writing vision (config: vision)](https://benchize.com/benchmarks/benchmark%3Avision%3Acreative_writing_vision--config-vision.md) Updated: 10/21/2026 Models tested: 70 Human pairwise preferences on creative writing vision prompts in Arena’s vision evaluation. The original organizer publishes model ratings and uncertainty; these are preference… Top models · Arena Latest result per model · Version not reported - 1[Gemini 3 Pro](https://benchize.com/models/gemini-3-pro.md) - 2[Claude Fable 5](https://benchize.com/models/claude-fable-5-high.md)High - 3[Claude Opus 4.6](https://benchize.com/models/claude-opus-4-6-high.md)High View details → ## [Arena vision style control / creative writing vision](https://benchize.com/benchmarks/lmarena%3Avision_style_control%3Acreative_writing_vision.md) Updated: 10/21/2026 Models tested: 70 Human pairwise preferences on creative writing vision prompts in Arena’s vision style control evaluation. The original organizer publishes model ratings and uncertainty, adjusting… Top models · Arena Latest result per model · Version not reported - 1[Gemini 3 Pro](https://benchize.com/models/gemini-3-pro.md) - 2[Claude Fable 5](https://benchize.com/models/claude-fable-5-high.md)High - 3[GPT-6.1 Sol](https://benchize.com/models/gpt-6-1-sol-max.md)Max View details → ## [Arena vision / chinese (config: vision)](https://benchize.com/benchmarks/benchmark%3Avision%3Achinese--config-vision.md) Updated: 10/21/2026 Models tested: 74 Human pairwise preferences on chinese prompts in Arena’s vision evaluation. The original organizer publishes model ratings and uncertainty; these are preference ratings rather… Top models · Arena Latest result per model · Version not reported - 1[Gemini 3.8 Flash](https://benchize.com/models/gemini-3-8-flash-high.md)High - 2[Claude Opus 5](https://benchize.com/models/claude-opus-5-high.md)High - 3[Gemini 3.7 Flash](https://benchize.com/models/gemini-3-7-flash-high.md)High View details → ## [Arena vision style control / chinese](https://benchize.com/benchmarks/lmarena%3Avision_style_control%3Achinese.md) Updated: 10/21/2026 Models tested: 74 Human pairwise preferences on chinese prompts in Arena’s vision style control evaluation. The original organizer publishes model ratings and uncertainty, adjusting for response… Top models · Arena Latest result per model · Version not reported - 1[Gemini 3.8 Flash](https://benchize.com/models/gemini-3-8-flash-high.md)High - 2[Gemini 3.7 Flash](https://benchize.com/models/gemini-3-7-flash-high.md)High - 3[Muse Spark 1.3](https://benchize.com/models/muse-spark-1-3-xhigh.md)Extra high View details → ## [Arena vision / humor (config: vision)](https://benchize.com/benchmarks/benchmark%3Avision%3Ahumor--config-vision.md) Updated: 10/21/2026 Models tested: 63 Human pairwise preferences on humor prompts in Arena’s vision evaluation. The original organizer publishes model ratings and uncertainty; these are preference ratings rather than… Top models · Arena Latest result per model · Version not reported - 1[Claude Fable 5](https://benchize.com/models/claude-fable-5-high.md)High - 2[Gemini 3 Pro](https://benchize.com/models/gemini-3-pro.md) - 3[Gemini 3.1 Pro Preview](https://benchize.com/models/gemini-3-1-pro-preview.md) View details → ## [Arena vision style control / humor](https://benchize.com/benchmarks/lmarena%3Avision_style_control%3Ahumor.md) Updated: 10/21/2026 Models tested: 63 Human pairwise preferences on humor prompts in Arena’s vision style control evaluation. The original organizer publishes model ratings and uncertainty, adjusting for response… Top models · Arena Latest result per model · Version not reported - 1[Claude Fable 5](https://benchize.com/models/claude-fable-5-high.md)High - 2[Gemini 3 Pro](https://benchize.com/models/gemini-3-pro.md) - 3[Gemini 3.1 Pro Preview](https://benchize.com/models/gemini-3-1-pro-preview.md) View details → ## [Arena vision / entity recognition (config: vision)](https://benchize.com/benchmarks/benchmark%3Avision%3Aentity_recognition--config-vision.md) Updated: 10/21/2026 Models tested: 34 Human pairwise preferences on entity recognition prompts in Arena’s vision evaluation. The original organizer publishes model ratings and uncertainty; these are preference ratings… Top models · Arena Latest result per model · Version not reported - 1[Gemini 3 Pro](https://benchize.com/models/gemini-3-pro.md) - 2[Gemini 3 Flash](https://benchize.com/models/gemini-3-flash.md) - 3[Gemini 3.1 Pro Preview](https://benchize.com/models/gemini-3-1-pro-preview.md) View details → ## [Arena vision style control / entity recognition](https://benchize.com/benchmarks/lmarena%3Avision_style_control%3Aentity_recognition.md) Updated: 10/21/2026 Models tested: 34 Human pairwise preferences on entity recognition prompts in Arena’s vision style control evaluation. The original organizer publishes model ratings and uncertainty, adjusting for… Top models · Arena Latest result per model · Version not reported - 1[Gemini 3 Pro](https://benchize.com/models/gemini-3-pro.md) - 2[Gemini 3 Flash](https://benchize.com/models/gemini-3-flash.md) - 3[Gemini 3.1 Pro Preview](https://benchize.com/models/gemini-3-1-pro-preview.md) View details → ## [Arena vision / captioning (config: vision)](https://benchize.com/benchmarks/benchmark%3Avision%3Acaptioning--config-vision.md) Updated: 10/21/2026 Models tested: 22 Human pairwise preferences on captioning prompts in Arena’s vision evaluation. The original organizer publishes model ratings and uncertainty; these are preference ratings rather… Top models · Arena Latest result per model · Version not reported - 1[Gemini 3.1 Pro Preview](https://benchize.com/models/gemini-3-1-pro-preview.md) - 2[Gemini 3 Pro](https://benchize.com/models/gemini-3-pro.md) - 3[Gemini 2.5 Pro](https://benchize.com/models/gemini-2-5-pro.md) View details → ## [Arena vision style control / captioning](https://benchize.com/benchmarks/lmarena%3Avision_style_control%3Acaptioning.md) Updated: 10/21/2026 Models tested: 22 Human pairwise preferences on captioning prompts in Arena’s vision style control evaluation. The original organizer publishes model ratings and uncertainty, adjusting for response… Top models · Arena Latest result per model · Version not reported - 1[Gemini 3.1 Pro Preview](https://benchize.com/models/gemini-3-1-pro-preview.md) - 2[Gemini 3 Pro](https://benchize.com/models/gemini-3-pro.md) - 3[GPT-5.2](https://benchize.com/models/gpt-5-2-high.md)High View details → ## [Arena vision / creative writing (config: vision)](https://benchize.com/benchmarks/benchmark%3Avision%3Acreative_writing--config-vision.md) Updated: 10/21/2026 Models tested: 13 Human pairwise preferences on creative writing prompts in Arena’s vision evaluation. The original organizer publishes model ratings and uncertainty; these are preference ratings… Top models · Arena Latest result per model · Version not reported - 1[Gemini 3 Pro](https://benchize.com/models/gemini-3-pro.md) - 2[Gemini 3 Flash](https://benchize.com/models/gemini-3-flash.md) - 3[GPT-5.1](https://benchize.com/models/gpt-5-1.md) View details → ## [Arena vision style control / creative writing](https://benchize.com/benchmarks/lmarena%3Avision_style_control%3Acreative_writing.md) Updated: 10/21/2026 Models tested: 13 Human pairwise preferences on creative writing prompts in Arena’s vision style control evaluation. The original organizer publishes model ratings and uncertainty, adjusting for… Top models · Arena Latest result per model · Version not reported - 1[Gemini 3 Pro](https://benchize.com/models/gemini-3-pro.md) - 2[Gemini 3 Flash](https://benchize.com/models/gemini-3-flash.md) - 3[GPT-5.1](https://benchize.com/models/gpt-5-1.md) View details → ## [VoiceCodeBench / ctem](https://benchize.com/benchmarks/vals-ai%3Avoice-code-bench%3Actem.md) Updated: 10/21/2026 Models tested: 1 Can speech recognition preserve exact structured values in workplace audio? Top models · Vals AI Latest result per model · 1 - 1[GPT-4o Transcribe](https://benchize.com/models/gpt-4o-transcribe.md) View details → ## [VoiceCodeBench / overall](https://benchize.com/benchmarks/vals-ai%3Avoice-code-bench%3Aoverall.md) Updated: 10/21/2026 Models tested: 1 Can speech recognition preserve exact structured values in workplace audio? Top models · Vals AI Latest result per model · 1 - 1[GPT-4o Transcribe](https://benchize.com/models/gpt-4o-transcribe.md) View details → ## [VoiceCodeBench / wer](https://benchize.com/benchmarks/vals-ai%3Avoice-code-bench%3Awer.md) Updated: 10/21/2026 Models tested: 1 Can speech recognition preserve exact structured values in workplace audio? Top models · Vals AI Latest result per model · 1 - 1[GPT-4o Transcribe](https://benchize.com/models/gpt-4o-transcribe.md) View details → ## [DesignArena Video editing](https://benchize.com/benchmarks/designarena%3Avideo-to-video.md) Updated: 10/21/2026 Models tested: 1 Human pairwise preference evaluation of video editing. Voters compare model outputs for the same request; published Elo ratings and separate model configurations are retained for… Top models · designarena Latest result per model · public - 1[Wan 2.7](https://benchize.com/models/wan-2-7.md) View details → ## Methodology and provenance Benchize collects and structures public benchmark reports. Scores, costs, timings, ranks and uncertainty remain attributed to their original sources. Different versions, configurations and measurement units are separate results. Missing fields stay unavailable. --- Cite as: Multimodal benchmarks — Benchize, https://benchize.com/benchmarks/category/multimodal (retrieved 2026-10-21)