AlibabaQwen3 235B A22B
Qwen3 235B A22B is an Alibaba open-weights reasoning model, released 28 Apr 2025. Its best configuration ranks #381 of 785 on the BenchLeader Index at 50.1 ±3.9, in the upper half. The ± is the point: 309 other configurations score within that range, so they and this one cannot be told apart on quality alone — price and speed are what separate them. It scores highest in maths (59) and lowest in instruction following (43). At $1.23 per million tokens blended it is pricier than most ranked models. It has been superseded by Qwen3 235B A22B 2507, which is the version to compare against today. It has been measured at 3 reasoning-effort settings; this summary describes the best-scoring one, and the tabs above switch between them. Last measured 22 Oct 2026.
- Blended price
- $1.23/M
- $0.700 in · $2.80 out
- Output speed
- –
- First answer
- –
- Context
- 131k
- Full answer
- –
- Cost per run
- –
- Released
- 28 Apr 2025
- Overall index50
- Reasoning46
- Coding49
- Maths59
- Knowledge50
- Instruction following43
- Human preference54
Versions
Alibaba has shipped 2 models under this name. Each is ranked on its own results; a newer version often has fewer results so far, which holds its index nearer the average until more arrive.
| Model | Released | Index | Rank |
|---|---|---|---|
| Qwen3 235B A22B 2507 | 22 Jul 2025 | 50.4 | #373 |
| Qwen3 235B A22Bthis page | 28 Apr 2025 | 50.1 | #381 |
Reasoning-effort configurations
The same model behaves differently depending on how much it is allowed to think. Each row is one setting, scored only on the benchmarks that were run at that setting. “Not stated” collects results from publishers that did not say which setting they used; for a reasoning model that is usually its thinking mode, but we do not assume it. Pick a setting here or at the top of the page to see its price, speed and category scores.
| Effort | Index | Rank | Speed | First answer | Chat reply cost | Categories |
|---|---|---|---|---|---|---|
| no reasoning | 45.4 | #514 | – | – | $0.0011 | Agents & tools 39 · Coding 57 · Composite 39 · Human preference 58 · Instruction following 40 · Knowledge 38 · Long context 23 · Maths 56 · Reasoning 48 |
| thinking | 42.3 | #601 | – | – | $0.0011 | Agents & tools 39 · Composite 40 · Instruction following 42 · Knowledge 41 · Long context 23 · Reasoning 48 |
| not statedbest | 50.1 | #381 | – | – | $0.0011 | Coding 49 · Human preference 54 · Instruction following 43 · Knowledge 50 · Maths 59 · Reasoning 46 |
How its index has moved
8 Sept 2026 to 21 Oct 2026- not stated
- no reasoning
- thinking
The index is recomputed from scratch every day, so a line moves when a new benchmark result lands, when a publisher revises a score, or when the models it is normalised against change. Early movement usually means the score is still settling.
Benchmark results
See all LLM benchmarks. One column per reasoning effort. Rank is among every configuration of every model on that benchmark. Hover a score for the run it came from.
Reasoning
| Benchmark | no reasoning | thinking | not stated | Source | Trend |
|---|---|---|---|---|---|
| GPQA Diamondeffort not stated | – | – | 70.7%#170 | Epoch AI Benchmarking Hub | |
| SimpleBencheffort not stated | – | – | 31.0%#67 | SimpleBench | |
| Arena Hard Prompts | 1420#146 | – | 1392#178 | Arena (LMArena) | |
| GPQA Diamond (AA)not in index | 61.3%#367 | 70.0%#298 | – | Artificial Analysis | |
| Humanity's Last Exam (AA)not in index | 4.2%#527 | 11.0%#304 | – | Artificial Analysis | |
| GPQA Diamond (Vals)not in indexeffort not stated | – | – | 70.2%#88 | Vals AI | |
| Kagi LLM Benchmark | 55.0%#66 | 69.4%#29 | – | Kagi LLM Benchmark | |
| GPQA Diamond (HELM)not in indexeffort not stated | – | – | 62.3%#24 | HELM Capabilities | |
| CritPt | 0.0%#315 | 0.0%#315 | – | Artificial Analysis | |
| SciPredictnot in indexeffort not stated | – | – | 16.6%#13 | Scale AI SEAL | |
| LMCAnot in indexeffort not stated | – | – | 25.0%#149 | Epoch AI Benchmarking Hub | |
| DTBenchnot in indexeffort not stated | – | – | 75.7%#147 | Epoch AI Benchmarking Hub | |
| ForecastBenchnot in indexeffort not stated | – | – | 64.1%#35 | ForecastBench |
Coding
| Benchmark | no reasoning | not stated | Source | Trend |
|---|---|---|---|---|
| WeirdMLeffort not stated | – | 37.3%#118 | WeirdML | |
| Arena Coding | 1444#150 | 1432#173 | Arena (LMArena) | |
| LiveCodeBencheffort not stated | – | 70.6%#84 | Vals AI | |
| SWE-Bench Proeffort not stated | – | 21.4%#18 | Scale AI SEAL | |
| Aider Polygloteffort not stated | – | 59.6%#16 | Aider polyglot leaderboard | |
| SWE-Bench Pro (private)not in indexeffort not stated | – | 21.4%#18 | Scale AI SEAL |
Agents & tools
| Benchmark | no reasoning | thinking | Source | Trend |
|---|---|---|---|---|
| Terminal-Bench Hard | 6.1%#266 | 6.1%#266 | Artificial Analysis | |
| τ²-Bench Telecom (AA)not in index | 27.2%#277 | 24.0%#303 | Artificial Analysis |
Maths
| Benchmark | no reasoning | not stated | Source | Trend |
|---|---|---|---|---|
| MATH Level 5effort not stated | – | 68.9%#40 | Epoch AI Benchmarking Hub | |
| AIME (Vals)effort not stated | – | 84.0%#40 | Vals AI | |
| MGSMeffort not stated | – | 92.5%#17 | Vals AI | |
| Omni-MATH (HELM)effort not stated | – | 54.8%#15 | HELM Capabilities | |
| MATH 500not in indexeffort not stated | – | 94.6%#8 | Vals AI | |
| Arena Maths | 1391#161 | 1392#160 | Arena (LMArena) |
Knowledge
| Benchmark | no reasoning | thinking | not stated | Source | Trend |
|---|---|---|---|---|---|
| AA-Omniscience | -52.4#409 | -44.7#350 | – | Artificial Analysis | |
| MMLU-Proeffort not stated | – | – | 81.3%#78 | Vals AI | |
| LegalBencheffort not stated | – | – | 80.2%#87 | Vals AI | |
| TaxEvaleffort not stated | – | – | 70.7%#85 | Vals AI | |
| MedQAeffort not stated | – | – | 90.6%#44 | Vals AI | |
| MultiNRCeffort not stated | – | – | 17.6%#37 | Scale AI SEAL | |
| MMLU-Pro (HELM)not in indexeffort not stated | – | – | 81.7%#15 | HELM Capabilities | |
| AA-Omniscience: accuracynot in index | 18.6%#332 | 18.5%#333 | – | Artificial Analysis | |
| AA-Omniscience: non-hallucinationnot in index | 12.9%#368 | 22.4%#265 | – | Artificial Analysis |
Instruction following
| Benchmark | no reasoning | thinking | not stated | Source | Trend |
|---|---|---|---|---|---|
| IFBench | 36.6%#299 | 38.7%#278 | – | Artificial Analysis | |
| MultiChallengeeffort not stated | – | – | 41.2%#30 | Scale AI SEAL | |
| IFEval (HELM)effort not stated | – | – | 81.6%#40 | HELM Capabilities | |
| WildBench (HELM)effort not stated | – | – | 82.8%#20 | HELM Capabilities | |
| Arena Instruction Followingnot in index | 1381#163 | – | 1357#184 | Arena (LMArena) |
Human preference
| Benchmark | no reasoning | not stated | Source | Trend |
|---|---|---|---|---|
| Arena Text | 1402#147 | 1375#177 | Arena (LMArena) | |
| Arena Creative Writingnot in index | 1366#150 | 1323#194 | Arena (LMArena) | |
| Arena Multi-turnnot in index | 1410#143 | 1369#180 | Arena (LMArena) | |
| Arena Longer Queriesnot in index | 1410#146 | 1378#180 | Arena (LMArena) | |
| Lech Mazur Creative Writingnot in indexeffort not stated | – | 83.0%#9 | Lech Mazur |
Long context
| Benchmark | no reasoning | thinking | Source | Trend |
|---|---|---|---|---|
| AA-LCR | 0.0%#493 | 0.0%#493 | Artificial Analysis |
Composite
| Benchmark | no reasoning | thinking | not stated | Source | Trend |
|---|---|---|---|---|---|
| Epoch Capabilities Indexnot in indexeffort not stated | – | – | 139.2#117 | Epoch AI Benchmarking Hub | |
| AA Intelligence Index v4.3.2 | 8.3#441 | 9.5#397 | – | Artificial Analysis | |
| HELM Capabilities meannot in indexeffort not stated | – | – | 72.6%#19 | HELM Capabilities |
Safety & honesty
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| FORTRESSnot in index | 61.4%#4 | Scale AI SEAL | |
| MASKnot in indexpossible contamination | 56.4%#43 | Scale AI SEAL |
Price history
The price above, per 1M tokens, every day it changed.
What a task costs
Estimates from list price, output speed and time to first answer for the best configuration. “With caching” assumes three-quarters of the input is served from the prompt cache. Reasoning tokens are not modelled.
| Workload | Tokens in / out | Cost | With caching | Time |
|---|---|---|---|---|
| Chat reply | 400 / 300 | $0.0011 | – | – |
| Summarize a 30-page report | 12,000 / 600 | $0.010 | – | – |
| Code edit | 6,000 / 1,500 | $0.0084 | – | – |
| Agentic coding session | 60,000 / 4,000 | $0.053 | – | – |
| Structured extraction | 2,000 / 200 | $0.0020 | – | – |
More from Alibaba
See also
Questions people ask
- What are Qwen3 235B A22B's benchmark scores?
- As of 24 Oct 2026, Qwen3 235B A22B scores 50.1 ±3.9 on the BenchLeader Index, strongest in maths (59) and weakest in instruction following (43). That rests on 22 independent benchmark results, and every score on this page links to its publisher.
- How does Qwen3 235B A22B rank?
- Qwen3 235B A22B is #381 of 785 ranked configurations on 24 Oct 2026. 309 other configurations score within its margin, so the exact position matters less than the band. The leader is Claude Opus 5.5 (max effort) at 72.6.
- How much does Qwen3 235B A22B cost per 1M tokens?
- As of 24 Oct 2026, Qwen3 235B A22B costs $0.700 per 1M input tokens and $2.80 per 1M output tokens, $1.23 blended (three input tokens per output token). The list price last changed on 24 Oct 2026.
- When was Qwen3 235B A22B released?
- Alibaba released Qwen3 235B A22B on 28 Apr 2025. BenchLeader has results for it at 3 reasoning-effort settings.
Data as of 24 Oct 2026. Compare these configurations.
Cite as: BenchLeader, “Qwen3 235B A22B: benchmarks, pricing, speed and rank”, data as of 24 Oct 2026, https://www.benchleader.com/models/qwen3-235b-a22b.