# Qwen3.8 2.4T A95B

> Source: Benchize (https://benchize.com) — https://benchize.com/models/qwen3-8-2-4t-a95b. When citing or reusing this content, credit "Benchize" and link to https://benchize.com/models/qwen3-8-2-4t-a95b.

Alibaba

Canonical URL: https://benchize.com/models/qwen3-8-2-4t-a95b
Last updated: Thu, 22 Oct 2026 05:40:11 GMT

Reproduction and quotation must name Benchize, link to the canonical HTML URL, and retain individual source citations.

Alibaba

## Benchmarks

| Benchmark | Score | ± | Rank | Cost | Source |  |
| --- | --- | --- | --- | --- | --- | --- |
| [ArXivLean 09/2026](https://benchize.com/benchmarks/matharena%3Aarxivlean--september.md) | 1.89% | 95% CI -1.77–5.55% | 6 / 6 | 9usd per problem | [matharena · arxivlean--september](https://matharena.ai/competition_tables/arxivlean--september) |  |
| ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 1.89 \| percent \| competition \| Higher is better \| \| `cost_per_problem` \| 9.00 \| usd_per_problem \| competition \| Lower is better \| ### Configuration competition arxivlean--september predicted false sourceModelName Qwen3.8-2.4T-A95B sourceModelVersion qwen/qwen3.8_2.4t_a95b ### Reported statistics Uncertainty 95% confidence interval: -1.77–5.55% confidenceInterval95 3.66 problems 53 ### Source [Original source](https://matharena.ai/competition_tables/arxivlean--september) Source matharena Original model label Qwen3.8-2.4T-A95B Source version arxivlean--september Observed 2026-10-22T05:30:05.069Z Evidence SHA-256 `5d42ba0bbc37cc684861959960e17e45edb8cd93395dfe74b44e3252f043c42d` Source locator `/table/rows/5` |  |  |  |  |  |  |
| [ArXivMath 09/2026](https://benchize.com/benchmarks/matharena%3Aarxiv--september.md) | 45.28% | 95% CI 35.8–54.76% | 5 / 8 | 2.16usd per problem | [matharena · arxiv--september](https://matharena.ai/competition_tables/arxiv--september) |  |
| ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 45.28 \| percent \| competition \| Higher is better \| \| `cost_per_problem` \| 2.16 \| usd_per_problem \| competition \| Lower is better \| ### Configuration competition arxiv--september predicted false sourceModelName Qwen3.8-2.4T-A95B sourceModelVersion qwen/qwen3.8_2.4t_a95b ### Reported statistics Uncertainty 95% confidence interval: 35.8–54.76% confidenceInterval95 9.48 problems 53 ### Source [Original source](https://matharena.ai/competition_tables/arxiv--september) Source matharena Original model label Qwen3.8-2.4T-A95B Source version arxiv--september Observed 2026-10-22T05:30:09.570Z Evidence SHA-256 `8973b7bb8ea675057eb4c03cedc79d40de49d0a1b15da24d387eaaad2d82b3be` Source locator `/table/rows/4` |  |  |  |  |  |  |
| [Creative Writing v3](https://benchize.com/benchmarks/eq-bench%3Acreative-writing-v3.md) | 1,842.5elo | — | 17 / 90 | — | [eq bench · creative-writing-v3](https://eqbench.com/creative_writing.js) |  |
| ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `elo_score` \| 1842.5 \| elo \| overall \| Higher is better \| \| `creative_writing_score` \| 16.72 \| rubric_score \| overall \| Higher is better \| ### Configuration interactiveAdjustments none protocol creative-writing-v3 sourceModelVersion Qwen/Qwen3.8-2.4T-A95B ### Source [Original source](https://eqbench.com/creative_writing.js) Source eq bench Original model label Qwen/Qwen3.8-2.4T-A95B Source version creative-writing-v3 Observed 2026-10-22T05:10:05.433Z Evidence SHA-256 `722ddb132e8d3a5e5c39b8d445f8475f3f7e24893a007260c6f3e3cce5c1ee18` Source locator `/leaderboardDataCreativeWritingV3/18` |  |  |  |  |  |  |
| [BrokenArXiv 09/2026](https://benchize.com/benchmarks/matharena%3Aarxiv_false--september.md) | 33.02% | 95% CI 24.07–41.97% | 5 / 8 | 2.69usd per problem | [matharena · arxiv_false--september](https://matharena.ai/competition_tables/arxiv_false--september) |  |
| ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 33.02 \| percent \| competition \| Higher is better \| \| `cost_per_problem` \| 2.69 \| usd_per_problem \| competition \| Lower is better \| ### Configuration competition arxiv_false--september predicted false sourceModelName Qwen3.8-2.4T-A95B sourceModelVersion qwen/qwen3.8_2.4t_a95b ### Reported statistics Uncertainty 95% confidence interval: 24.07–41.97% confidenceInterval95 8.95 problems 53 ### Source [Original source](https://matharena.ai/competition_tables/arxiv_false--september) Source matharena Original model label Qwen3.8-2.4T-A95B Source version arxiv_false--september Observed 2026-10-22T05:30:07.069Z Evidence SHA-256 `4431a852d035438c3451a917deda38fbe7f7e38a2da0903d97e836c3b8776de1` Source locator `/table/rows/4` |  |  |  |  |  |  |

---
Cite as: Qwen3.8 2.4T A95B — Benchize, https://benchize.com/models/qwen3-8-2-4t-a95b (retrieved 2026-10-22)