Conquered benchmarks
Benchmarks where at least one model has achieved a reported score of exactly 100%.
Published
Terminal-Bench 4.0 / security
ConqueredUpdated:
Models tested: 36
Frontier-difficulty terminal tasks across software, science, ML, operations, hardware, security, and media
Top models · Vals AI
Latest result per model · 4.0
- 1Claude Opus 5.5Max100%
- 2Claude Opus 5Max100%
- 3Claude Sonnet 5.5Max93.33%
Top models · Vals AI
Latest result per model · 1.1
- 1Claude Fable 5.1Max100%
- 2Claude Sonnet 5.5Max100%
- 3Claude Opus 5.5Max100%
OTIS Mock AIME 2024–2025
ConqueredUpdated:
Models tested: 123
A collection of 45 unofficial mock AIME mathematics problems from the OTIS 2024 exam and the two 2025 exams. Each problem has an integer answer between 0 and 999; models’…
Top models · epoch ai
Latest result per model · OTIS Mock AIME 2024-2025
- 1GPT-6.1 SolMax100%
- 2Claude Fable 5.1Max100%
- 3GPT-5.6 SolMax100%
Top models · Vals AI
Latest result per model · 2
- 1GPT-6 SolMax100%
- 2Claude Fable 5.1Max100%
- 3Gemini 4 ArgonHigh100%
Top models · Vals AI
Latest result per model · 2
- 1GPT-6 AstraMax100%
- 2GPT-6.1 SolMax100%
- 3Gemini 4 ArgonHigh100%
Top models · Vals AI
Latest result per model · 2
- 1GPT-5.6 TerraMax100%
- 2GPT-6 AstraMax100%
- 3Gemini 4 ArgonHigh100%
Top models · Vals AI
Latest result per model · 2
- 1Gemini 4 ArgonHigh100%
- 2GPT-6 AstraMax100%
- 3GPT-6.1 SolMax96.89%
Terminal-Bench 2.1 / easy
ConqueredUpdated:
Models tested: 43
State-of-the-art set of difficult terminal-based tasks
Top models · Vals AI
Latest result per model · 2.1
- 1Claude Sonnet 4.6Non-thinking100%
- 2Claude Fable 5Max100%
- 3GPT-5.6 TerraMax100%
FrontierMath Tier 4 v2 (private)
ConqueredUpdated:
Models tested: 38
The corrected v2 private Tier 4 subset of FrontierMath, containing exceptionally difficult research-level mathematics problems. Models may reason with Python and submit…
Top models · epoch ai
Latest result per model · FrontierMath-Tier-4-v2-Private
- 1GPT-6.1 SolMax100%
- 2GPT-6 AstraExtra high97.6%
- 3GPT-6 AstraMax97.6%
BrokenArXiv 06/2026
ConqueredUpdated:
Models tested: 21
BrokenArXiv 06/2026: Recognizing false mathematical claims derived from recent research papers instead of producing invalid proofs. This competition contains 54 problems.…
Top models · matharena
Latest result per model · arxiv_false--june
- 1Claude Opus 5.5High100%
- 2GPT-6.1 SolMax100%
- 3GPT-6 AstraMax99.07%
SWE-bench / 1-4 hours
ConqueredUpdated:
Models tested: 39
Solving production software engineering tasks
Top models · Vals AI
Latest result per model · 1
- 1DeepSeek V4 Pro 0813Max100%
- 2GPT-5.6 TerraMax97.62%
- 3GPT-5.6 SolMax97.62%
SWE-bench / >4 hours
ConqueredUpdated:
Models tested: 39
Solving production software engineering tasks
Top models · Vals AI
Latest result per model · 1
- 1GPT-5.6 SolMax100%
- 2Claude Fable 5Max100%
- 3DeepSeek V4 Pro 0813Max100%
ClaimBench / direct support
ConqueredUpdated:
Models tested: 5
Does a 10-K excerpt support a claim, contradict it, or say nothing about it?
Top models · Vals AI
Latest result per model · 1
- 1Claude Sonnet 5.5Max100%
- 2GPT-6 LunaMax100%
- 3Claude Haiku 5.5Max100%
ClaimBench / missing evidence
ConqueredUpdated:
Models tested: 5
Does a 10-K excerpt support a claim, contradict it, or say nothing about it?
Top models · Vals AI
Latest result per model · 1
- 1Claude Haiku 5.5Max100%
- 2GPT-6 LunaMax100%
- 3Claude Opus 5.5Max100%
ClaimBench / non supporting citation
ConqueredUpdated:
Models tested: 5
Does a 10-K excerpt support a claim, contradict it, or say nothing about it?
Top models · Vals AI
Latest result per model · 1
- 1Claude Opus 5.5Max100%
- 2GPT-6.1 SolMax100%
- 3GPT-6 LunaMax100%
ClaimBench / numeric discrepancy
ConqueredUpdated:
Models tested: 5
Does a 10-K excerpt support a claim, contradict it, or say nothing about it?
Top models · Vals AI
Latest result per model · 1
- 1GPT-6.1 SolMax100%
- 2GPT-6 LunaMax100%
- 3Claude Opus 5.5Max100%
ClaimBench / overall
ConqueredUpdated:
Models tested: 5
Does a 10-K excerpt support a claim, contradict it, or say nothing about it?
Top models · Vals AI
Latest result per model · 1
- 1Claude Opus 5.5Max100%
- 2Claude Sonnet 5.5Max100%
- 3GPT-6.1 SolMax100%
ClaimBench / subtle contradiction
ConqueredUpdated:
Models tested: 5
Does a 10-K excerpt support a claim, contradict it, or say nothing about it?
Top models · Vals AI
Latest result per model · 1
- 1Claude Haiku 5.5Max100%
- 2Claude Sonnet 5.5Max100%
- 3Claude Opus 5.5Max100%
Robo-bench / balance scale
ConqueredUpdated:
Models tested: 5
Can a model control a robot arm and a humanoid in a real-time physics simulation, from motor commands and camera images alone? Most tasks are held out.
Top models · Vals AI
Latest result per model · 1
- 1GPT-6.1 SolMax100%
- 2Claude Opus 5.5Max95%
- 3Claude Fable 5.1Max10%
Terminal-Bench 2.0 / easy
ConqueredUpdated:
Models tested: 23
State-of-the-art set of difficult terminal-based tasks
Top models · Vals AI
Latest result per model · 2.0
- 1Claude Opus 4.7High100%
- 2Gemini 2.5 Pro100%
- 3Gemini 3 Flash PreviewHigh100%
AIME 2026
ConqueredUpdated:
Models tested: 21
AIME 2026: Solving final-answer mathematics competition problems. This competition contains 30 problems. MathArena reports normalized performance across repeated attempts and a…
Top models · matharena
Latest result per model · aime--aime_2026
- 1GPT-5.5Extra high100%
- 2Claude Opus 4.8Max100%
- 3GPT-5.4Extra high99.17%
Kangaroo 2025 11-12
ConqueredUpdated:
Models tested: 16
Kangaroo 2025 11-12: Mathematical reasoning on school-level multiple-choice problems, including visual diagrams and patterns. This competition contains 30 problems. MathArena…
Top models · matharena
Latest result per model · kangaroo--kangaroo_2025_11-12
- 1Claude Opus 4.8Max100%
- 2GPT-5.4Extra high98.33%
- 3GPT-5.5Extra high98.33%
Kangaroo 2025 9-10
ConqueredUpdated:
Models tested: 15
Kangaroo 2025 9-10: Mathematical reasoning on school-level multiple-choice problems, including visual diagrams and patterns. This competition contains 30 problems. MathArena…
Top models · matharena
Latest result per model · kangaroo--kangaroo_2025_9-10
- 1GPT-5.2High100%
- 2Gemini 3.6 Flash100%
- 3Gemini 3.5 Flash100%
AIME 2025
ConqueredUpdated:
Models tested: 49
AIME 2025: Solving final-answer mathematics competition problems. This competition contains 30 problems. MathArena reports normalized performance across repeated attempts and a…
Top models · matharena
Latest result per model · aime--aime_2025
- 1GPT-5.2High100%
- 2GPT-5.2Extra high100%
- 3Step 3.5 Flash98.33%
HMMT Feb 2025
ConqueredUpdated:
Models tested: 34
HMMT Feb 2025: Solving final-answer mathematics competition problems. This competition contains 30 problems. MathArena reports normalized performance across repeated attempts and…
Top models · matharena
Latest result per model · hmmt--hmmt_feb_2025
- 1GPT-5.2Extra high100%
- 2Step 3.5 Flash98.33%
- 3GPT-5.2High98.33%
BRUMO 2025
ConqueredUpdated:
Models tested: 25
BRUMO 2025: Solving final-answer mathematics competition problems. This competition contains 30 problems. MathArena reports normalized performance across repeated attempts and a…
Top models · matharena
Latest result per model · brumo--brumo_2025
- 1Step 3.5 Flash100%
- 2Gemini 3 Flash100%
- 3GLM-599.17%
MGSM / mgsm en
ConqueredUpdated:
Models tested: 8
A multilingual benchmark for mathematical questions.
Top models · Vals AI
Latest result per model · 1
- 1MiniMax M2.1100%
- 2Gemini 3 Flash PreviewHigh98.8%
- 3Mistral Large 3 251298.4%
DiploBench exploratory games
ConqueredUpdated:
Models tested: 6
Exploratory Diplomacy games testing negotiation and strategic reasoning while the evaluated model plays Austria against Gemini opponents. Each game remains separate; the author…
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.