Rankings / Alibaba
Qwen3 235B A22b 2507
released 2025-07-25
BenchAtlas Index
as of 2026-07-14
thinking
rank #275
11 families · 5 categories · high
Benchmark evidence
19 results
agentic coding
| τ²-Bench | thinking | 53.2 % | independent | Artificial Analysis |
| τ²-Bench subset=banking | thinking | 7.8 % | independent | Artificial Analysis |
coding
| LiveCodeBench v6 implementation=artificial-analysis | thinking | 78.8 % | independent | Artificial Analysis |
| SciCode | thinking | 42.4 % | independent | Artificial Analysis |
external indices
| AA Coding Index | thinking | 22.1 points | independent | Artificial Analysis |
| Intelligence Index v4.1 | thinking | 19.6 points | independent | Artificial Analysis |
| AA Math Index | thinking | 91.0 points | independent | Artificial Analysis |
| ECI | thinking | 144.8 points | independent | Epoch AI Benchmarking Hub |
factuality
| SimpleQA Verified | thinking | 50.1 % | independent | Epoch AI Benchmarking Hub 2025-12-11 |
knowledge science
| GPQA Diamond | thinking | 80.0 % | independent | Epoch AI Benchmarking Hub 2025-12-11 |
| GPQA Diamond implementation=artificial-analysis | thinking | 79.0 % | independent | Artificial Analysis |
| Humanity's Last Exam implementation=artificial-analysis | thinking | 15.0 % | independent | Artificial Analysis |
| MMLU-Pro implementation=artificial-analysis | thinking | 84.3 % | independent | Artificial Analysis |
long context instruction
| AA-LCR | thinking | 67.0 % | independent | Artificial Analysis |
| IFBench | thinking | 51.2 % | independent | Artificial Analysis |
reasoning math
| AIME implementation=artificial-analysis | thinking | 94.0 % | independent | Artificial Analysis |
| AIME year=2025 · implementation=artificial-analysis | thinking | 91.0 % | independent | Artificial Analysis |
| MATH-500 implementation=artificial-analysis | thinking | 98.4 % | independent | Artificial Analysis |
| OTIS Mock AIME 2024–2025 | thinking | 86.7 % | independent | Epoch AI Benchmarking Hub 2025-12-10 |
Agent + model results
systems, not bare-model scores
| agent + model Artificial Analysis harness + Qwen3 235B A22b 2507 | Terminal-Bench Hard | 13.6 % | independent | Artificial Analysis |
| agent + model Artificial Analysis harness + Qwen3 235B A22b 2507 | Terminal-Bench 2.1 | 12.0 % | independent | Artificial Analysis |
These scores measure the whole agent system (scaffold, tools, budgets) — they are never merged into the bare model’s numbers.