Qwen3.8 Max (0902) Benchmarks, Specifications & Availability
Explore Qwen3.8 Max (0902) from Alibaba (Qwen): published specifications, source-linked vendor benchmarks, independent evaluator coverage and recorded pricing when available.
Compare Qwen3.8 Max (0902) with other models →Explore data coverage
Published specifications
- Provider
- Alibaba (Qwen)
- Access
- Open weights
- License
- Qwen3.8-Max License
- Context window
- 1M
- Total parameters
- 2.4T
- Active parameters
- 95B
- Released
- 2026-09-02
- Modalities
- text, image, video
- Family
- Qwen3.8
Model card · Announcement · Website · OpenRouter
Model notes
Post-training upgrade of Qwen3.8-Max focused on coding/agents (0902 snapshot). Open weights: Qwen/Qwen3.8-2.4T-A95B on HF, ungated, Qwen3.8-Max License (checked 2026-09-28); README confirms 2.4T total / 95B activated.
Pricing · OpenRouter
Dated cached OpenRouter rates in USD per 1M tokens. Open the dashboard for live enhancements. Per-metric endpoint minima can refer to different providers; they are not a guaranteed combined rate from one endpoint.
| Tier | Input / 1M | Output / 1M | Cached input / 1M | Cache write / 1M | Date & source |
|---|---|---|---|---|---|
| Default | [object Object] | [object Object] | [object Object] | [object Object] | 2026-10-03 · OpenRouter source |
Recorded pricing notes
min-healthy endpoint minima 2026-10-01; min-healthy endpoint minima 2026-10-01; min-healthy endpoint minima 2026-10-02; min-healthy endpoint minima 2026-10-03
Official / vendor benchmarks
Default headline records. Own-vendor, peer-vendor and third-party provenance remain visible in evidence; configurations may differ.
| Benchmark / evaluator | Headline score | Evidence |
|---|---|---|
| CoWorkBench | 76.1%unknown effort | All 1 recorded result & sources76.1% · raw 76.1 % Headline · unknown effort · Third-party Source/record date: 2026-09-02 Alibaba published comparison as reported by DataCamp https://www.datacamp.com/blog/qwen3-8-max |
| DeepSWE v1.1 | 69.3%unknown effort | All 3 recorded results & sources69.3% · raw 69.3 % Headline · unknown effort · Third-party Source/record date: 2026-09-02 Alibaba published comparison as reported by DataCamp https://www.datacamp.com/blog/qwen3-8-max69.3% · raw 69.3 % Alternative · unknown effort · Peer vendor Source/record date: 2026-09-08 As reported in Nex-N2.5-Pro model card comparison table | Demoted 2026-09-25: duplicate of headline from Alibaba Qwen3.8-Max-0902 comparison table via DataCamp; vendor official preferred over Nex-N2.5-Pro peer comparison table (rule a); same value. https://huggingface.co/nex-agi/Nex-N2.5-Pro56.6% · raw 56.6 % Alternative · unknown effort · Peer vendor Source/record date: 2026-09-27 As reported by NaiveAI. Chart labels Qwen-3.8-Max; attached to the 0902 snapshot with version ambiguity noted. https://naive.ai/en/research/ |
| GPQA Diamond | 92.6%unknown effort | All 1 recorded result & sources92.6% · raw 92.6 % Headline · unknown effort · Third-party Source/record date: 2026-09-02 GPQA Diamond https://www.datacamp.com/blog/qwen3-8-max |
| HealthBench | 60.2%unknown effort | All 1 recorded result & sources60.2% · raw 60.2 % Headline · unknown effort · Third-party Source/record date: 2026-09-02 HealthBench https://www.datacamp.com/blog/qwen3-8-max |
| Humanity's Last Exam | 43.6%unknown effort | All 1 recorded result & sources43.6% · raw 43.6 % Headline · unknown effort · Third-party Source/record date: 2026-09-02 Humanity's Last Exam https://www.datacamp.com/blog/qwen3-8-max |
| IFBench | 82.8%unknown effort | All 1 recorded result & sources82.8% · raw 82.8 % Headline · unknown effort · Third-party Source/record date: 2026-09-02 IFBench instruction following https://www.datacamp.com/blog/qwen3-8-max |
| JobBench | 64%unknown effort | All 2 recorded results & sources64% · raw 64 % Headline · unknown effort · Third-party Source/record date: 2026-09-02 Alibaba published comparison as reported by DataCamp https://www.datacamp.com/blog/qwen3-8-max53.4% · raw 53.4 % Alternative · unknown effort · Peer vendor Source/record date: 2026-09-08 As reported in Nex-N2.5-Pro model card comparison table | Demoted 2026-09-25: duplicate of headline from Alibaba Qwen3.8-Max-0902 comparison table via DataCamp (64.0); vendor official 0902 figure preferred (rules a+b). 53.4 equals the pre-0902 Qwen3.8-Max (2026-08-03) value in Alibaba's table, so Nex's "Qwen3.8-Max" column appears to be the original snapshot, not 0902. https://huggingface.co/nex-agi/Nex-N2.5-Pro |
| LVBench | 81.8%max effort | All 1 recorded result & sources81.8% · raw 81.8 % Headline · max effort · Third-party Source/record date: 2026-09-02 LVBench — Qwen3.8-Max family table https://www.datacamp.com/blog/qwen3-8-max |
| MLS-Bench-Lite | 50.1%unknown effort | All 1 recorded result & sources50.1% · raw 50.1 % Headline · unknown effort · Third-party Source/record date: 2026-09-02 MLS-Bench-Lite https://www.datacamp.com/blog/qwen3-8-max |
| MobileWorld | 77.8%unknown effort | All 1 recorded result & sources77.8% · raw 77.8 % Headline · unknown effort · Third-party Source/record date: 2026-09-02 MobileWorld https://www.datacamp.com/blog/qwen3-8-max |
| NL2Repo-Bench | 64.9%unknown effort | All 2 recorded results & sources64.9% · raw 64.9 % Headline · unknown effort · Third-party Source/record date: 2026-09-02 Alibaba published comparison as reported by DataCamp https://www.datacamp.com/blog/qwen3-8-max55.9% · raw 55.9 % Alternative · unknown effort · Peer vendor Source/record date: 2026-09-27 As reported by NaiveAI. Chart labels Qwen-3.8-Max; attached to the 0902 snapshot with version ambiguity noted. https://naive.ai/en/research/ |
| OSWorld-Verified | 86.1%unknown effort | All 2 recorded results & sources86.1% · raw 86.1 % Headline · unknown effort · Third-party Source/record date: 2026-09-02 OSWorld-Verified https://www.datacamp.com/blog/qwen3-8-max86.1% · raw 86.1 % Alternative · unknown effort · Peer vendor Source/record date: 2026-09-08 As reported in Nex-N2.5-Pro model card comparison table | Demoted 2026-09-25: duplicate of headline from Alibaba Qwen3.8-Max-0902 comparison table via DataCamp; vendor official preferred over Nex-N2.5-Pro peer comparison table (rule a); same value. https://huggingface.co/nex-agi/Nex-N2.5-Pro |
| PaperBench | 93%max effort | All 1 recorded result & sources93% · raw 93 % Headline · max effort · Third-party Source/record date: 2026-09-02 PaperBench — DataCamp table for Qwen3.8-Max (family); confirm if 0902-identical https://www.datacamp.com/blog/qwen3-8-max |
| PerceptionBench | 63.5%max effort | All 1 recorded result & sources63.5% · raw 63.5 % Headline · max effort · Third-party Source/record date: 2026-09-02 PerceptionBench — Qwen3.8-Max family table https://www.datacamp.com/blog/qwen3-8-max |
| ProgramBench (Almost Solved) | 28%unknown effort | All 1 recorded result & sources28% · raw 28 % Headline · unknown effort · Third-party Source/record date: 2026-09-02 ProgramBench Almost Solved; Alibaba 0902 table via DataCamp https://www.datacamp.com/blog/qwen3-8-max |
| QwenSWEBench V2 | 70%unknown effort | All 1 recorded result & sources70% · raw 70 % Headline · unknown effort · Third-party Source/record date: 2026-09-02 Alibaba published comparison as reported by DataCamp https://www.datacamp.com/blog/qwen3-8-max |
| SWE-Atlas QnA | 66.3%unknown effort | All 1 recorded result & sources66.3% · raw 66.3 % Headline · unknown effort · Third-party Source/record date: 2026-09-02 SWE-Atlas QnA https://www.datacamp.com/blog/qwen3-8-max |
| SWE-bench Pro | 67.7%unknown effort | All 3 recorded results & sources67.7% · raw 67.7 % Headline · unknown effort · Third-party Source/record date: 2026-09-02 SWE-Pro / SWE-bench Pro https://www.datacamp.com/blog/qwen3-8-max67.7% · raw 67.7 % Alternative · unknown effort · Peer vendor Source/record date: 2026-09-08 As reported in Nex-N2.5-Pro model card comparison table | Demoted 2026-09-25: duplicate of headline from Alibaba Qwen3.8-Max-0902 comparison table via DataCamp; vendor official preferred over Nex-N2.5-Pro peer comparison table (rule a); same value. https://huggingface.co/nex-agi/Nex-N2.5-Pro67.7% · raw 67.7 % Alternative · unknown effort · Peer vendor Source/record date: 2026-09-27 As reported by NaiveAI. Chart labels Qwen-3.8-Max; attached to the 0902 snapshot with version ambiguity noted. https://naive.ai/en/research/ |
| SWE-Marathon | 44.8%unknown effort | All 1 recorded result & sources44.8% · raw 44.8 % Headline · unknown effort · Third-party Source/record date: 2026-09-02 SWE-Marathon https://www.datacamp.com/blog/qwen3-8-max |
| Terminal-Bench 2.1 | 86.6%max effort | All 3 recorded results & sources86.6% · raw 86.6 % Headline · max effort · Third-party Source/record date: 2026-09-02 Terminal-Bench 2.1 — Qwen3.8-Max family table https://www.datacamp.com/blog/qwen3-8-max86.6% · raw 86.6 % Alternative · unknown effort · Peer vendor Source/record date: 2026-09-08 As reported in Nex-N2.5-Pro model card comparison table | Demoted 2026-09-25: duplicate of headline from Alibaba Qwen3.8-Max-0902 comparison table via DataCamp; vendor official preferred over Nex-N2.5-Pro peer comparison table (rule a); same value (Nex row effort unknown, vendor row max). https://huggingface.co/nex-agi/Nex-N2.5-Pro88.8% · raw 88.8 % Alternative · unknown effort · Peer vendor Source/record date: 2026-09-27 As reported by NaiveAI. Chart labels Qwen-3.8-Max; attached to the 0902 snapshot with version ambiguity noted. https://naive.ai/en/research/ |
| Terminal-Bench 3.0 | 29%unknown effort | All 1 recorded result & sources29% · raw 29 % Headline · unknown effort · Third-party Source/record date: 2026-09-02 Alibaba published comparison as reported by DataCamp https://www.datacamp.com/blog/qwen3-8-max |
| Toolathlon-Verified | 73.3%unknown effort | All 2 recorded results & sources73.3% · raw 73.3 % Headline · unknown effort · Third-party Source/record date: 2026-09-02 Alibaba published comparison as reported by DataCamp https://www.datacamp.com/blog/qwen3-8-max72.5% · raw 72.5 % Alternative · unknown effort · Peer vendor Source/record date: 2026-09-08 As reported in Nex-N2.5-Pro model card comparison table | Demoted 2026-09-25: duplicate of headline from Alibaba Qwen3.8-Max-0902 comparison table via DataCamp (73.3); vendor official 0902 figure preferred (rules a+b). 72.5 equals the pre-0902 Qwen3.8-Max (2026-08-03) value in Alibaba's table, so Nex's "Qwen3.8-Max" column appears to be the original snapshot, not 0902. https://huggingface.co/nex-agi/Nex-N2.5-Pro |
| Vision2Web | 69%unknown effort | All 2 recorded results & sources69% · raw 69 % Headline · unknown effort · Third-party Source/record date: 2026-09-02 Vision2Web https://www.datacamp.com/blog/qwen3-8-max75.1% · raw 75.1 % Alternative · unknown effort · Peer vendor Source/record date: 2026-09-08 As reported in Nex-N2.5-Pro model card comparison table | Harness: Nex-AGI Vision2Web eval (avg of Frontend/Webpage/Website; Gemini-3.5-Flash VLM judge, GLM-5V-Turbo (Claude Code) GUI agent; Nex card footnote 7) — different harness from Alibaba's figure. Demoted 2026-09-25: duplicate of headline from Alibaba Qwen3.8-Max table via DataCamp (69.0); vendor official preferred (rule a). https://huggingface.co/nex-agi/Nex-N2.5-Pro |
| WorkArena | 1468unknown effort | All 1 recorded result & sources1468 · raw 1468 Elo Headline · unknown effort · Third-party Source/record date: 2026-09-02 Elo; Alibaba via DataCamp https://www.datacamp.com/blog/qwen3-8-max |
| AutomationBench v1.0.6 | 39.8%unknown effort | All 1 recorded result & sources39.8% · raw 39.8 % Headline · unknown effort · Peer vendor Source/record date: 2026-09-08 As reported in Nex-N2.5-Pro model card comparison table https://huggingface.co/nex-agi/Nex-N2.5-Pro |
| GDPval-AA v2 | 1717unknown effort | All 1 recorded result & sources1717 · raw 1717 Elo Headline · unknown effort · Peer vendor Source/record date: 2026-09-08 As reported in Nex-N2.5-Pro model card comparison table https://huggingface.co/nex-agi/Nex-N2.5-Pro |
| OSWorld 2.0 | 46.7%unknown effort | All 1 recorded result & sources46.7% · raw 46.7 % Headline · unknown effort · Peer vendor Source/record date: 2026-09-08 As reported in Nex-N2.5-Pro model card comparison table https://huggingface.co/nex-agi/Nex-N2.5-Pro |
| WebTest | 52.3%unknown effort | All 1 recorded result & sources52.3% · raw 52.3 % Headline · unknown effort · Peer vendor Source/record date: 2026-09-08 As reported in Nex-N2.5-Pro model card comparison table https://huggingface.co/nex-agi/Nex-N2.5-Pro |
| WebArena-Verified | 66.8%unknown effort | All 1 recorded result & sources66.8% · raw 66.8 % Headline · unknown effort · Peer vendor Source/record date: 2026-09-08 As reported in Nex-N2.5-Pro model card comparison table https://huggingface.co/nex-agi/Nex-N2.5-Pro |
| OSWorld-G | 84.9%unknown effort | All 1 recorded result & sources84.9% · raw 84.9 % Headline · unknown effort · Peer vendor Source/record date: 2026-09-08 As reported in Nex-N2.5-Pro model card comparison table https://huggingface.co/nex-agi/Nex-N2.5-Pro |
| SWE-MM | 39.2%unknown effort | All 1 recorded result & sources39.2% · raw 39.2 % Headline · unknown effort · Peer vendor Source/record date: 2026-09-08 As reported in Nex-N2.5-Pro model card comparison table https://huggingface.co/nex-agi/Nex-N2.5-Pro |
| OmniDoc | 92.1%unknown effort | All 1 recorded result & sources92.1% · raw 92.1 % Headline · unknown effort · Peer vendor Source/record date: 2026-09-08 As reported in Nex-N2.5-Pro model card comparison table https://huggingface.co/nex-agi/Nex-N2.5-Pro |
| MMMU Pro (no tools) | 82.7%unknown effort | All 1 recorded result & sources82.7% · raw 82.7 % Headline · unknown effort · Peer vendor Source/record date: 2026-09-23 As reported from Alibaba Qwen3.8-Max-0902 published comparison table (DataCamp transcription); MMMU-Pro https://www.datacamp.com/blog/qwen3-8-max |
| FrontierSWE | 73.5%unknown effort | All 1 recorded result & sources73.5% · raw 73.5 % Headline · unknown effort · Peer vendor Source/record date: 2026-09-27 As reported by NaiveAI. Chart labels Qwen-3.8-Max; attached to the 0902 snapshot with version ambiguity noted. https://naive.ai/en/research/ |
Independent evaluators
Evaluator harnesses are distinct from vendor measurements. Missing coverage is not a failed test.
| Benchmark / evaluator | Headline score | Evidence |
|---|---|---|
| Intelligence Index · Artificial Analysis | 45max effort | All 1 recorded result & sources45 · raw 45 index Headline · max effort · Independent evaluator Source/record date: 2026-10-03 Current AA model leaderboard; https://artificialanalysis.ai/models/qwen3-8-max |
| Cost per Intelligence Index task · Artificial Analysis | $5.41max effort | All 1 recorded result & sources$5.41 · raw 5.41 USD Headline · max effort · Independent evaluator Source/record date: 2026-10-03 Current AA model leaderboard; https://artificialanalysis.ai/models/qwen3-8-max |
| Output speed · Artificial Analysis | 39 tok/smax effort | All 1 recorded result & sources39 tok/s · raw 39 tok/s Headline · max effort · Independent evaluator Source/record date: 2026-10-03 Current AA model leaderboard; Median output tokens/s; leaderboard rounds to whole tokens. https://artificialanalysis.ai/models/qwen3-8-max |
| Vals Index · Vals AI | 51.8%max effort | All 1 recorded result & sources51.8% · raw 51.84 % Headline · max effort · Independent evaluator Source/record date: 2026-09-23 Vals lists alibaba/qwen3.8-max (not an explicit 0902 slug); mapped to site model qwen3.8-max-0902 https://www.vals.ai/benchmarks/vals_index |
| Bugs fixed /105 · Bug Hunt Bench | 25.7 fixesmax effort | All 1 recorded result & sources25.7 fixes · raw 25.7 fixes Headline · max effort · Independent evaluator Source/record date: 2026-09-23 Claude Code / Alibaba API max; 3-run mean — listed as Qwen3.8-Max; board data/benchmark.json updated 2026-09-23 https://github.com/phuryn/bug-hunt-bench |
| Hard board % of roofline · KernelBench (community board) | 24.1%max effort | All 1 recorded result & sources24.1% · raw 24.1 % Headline · max effort · Independent evaluator Source/record date: 2026-09-23 Hard 5/6 on qwen3.8-max board; CUDA 19.3% 2/4; some cells audit-flagged https://kernelbench.com/models/qwen3.8-max |
| GDPval-AA Elo · Artificial Analysis | 1668unknown effort | All 1 recorded result & sources1668 · raw 1668 Elo Headline · unknown effort · Independent evaluator Source/record date: 2026-09-23 GDPval-AA v2.1 Elo; Qwen3.8 Max (0902) https://artificialanalysis.ai/evaluations/gdpval-aa |
| CUDA board % of roofline · KernelBench (community board) | 19.3%max effort | All 1 recorded result & sources19.3% · raw 19.3 % Headline · max effort · Independent evaluator Source/record date: 2026-09-23 CUDA 2/4; Hard 24.1% already ingested; some cells audit-flagged https://kernelbench.com/models/qwen3.8-max |
Read how we select and source scores or the comparison guide.