Qwen3.8-Flash-Next
Qwen3.8-Flash-NextCategory scores
Scores per capability area with the model's rank among ranked models in that category.
- 1Instruction following#2291.2
- 2Multimodal#887.7
- 3Reasoning79.8
- 4Coding#4258.6
- 5Knowledge#5358.1
- 6Agentic & tool use40.0
Benchmark results (24)
As reported in system cards, papers and public evaluations. Position compares this model with every other model reporting the same benchmark.
Multimodal
| CharXiv | 90.6 | #6 of 42 |
| ExtractBench (LlamaIndex)(lower is better) | 89.88 | #16 of 24 |
| MathVision | 90.6 | #5 of 20 |
| ERQA | 72.3 | #2 of 16 |
| RealWorldQA | 88.5 | #1 of 16 |
| CharXiv w/o tools | 84.6 | #5 of 14 |
| LVBench | 76.6 | #6 of 6 |
| MathVision w/ Python | 95.7 | #4 of 5 |
| Vision2Web | 64.0 | #2 of 4 |
Agentic & tool use
| JobBench | 55.7 | #10 of 34 |
| OSWorld 2.0 | 19.4 | #15 of 24 |
| Toolathlon-Verified | 73.5 | #11 of 22 |
| Agents' Last Exam | 51.2 | #4 of 17 |
| AndroidWorld | 84.5 | #3 of 8 |
| CoWorkBench | 73.9 | #3 of 4 |
Coding
| SWE-bench Pro | 62.5 | #20 of 69 |
| DeepSWE | 58.7 | #38 of 50 |
| LiveCodeBench v6 | 91.9 | #3 of 46 |
| NL2Repo | 48.1 | #14 of 31 |
Knowledge
| GPQA | 91.7 | #17 of 117 |
| HLE | 35.9 | #35 of 69 |
| HLE w/o tools | 35.9 | #26 of 43 |
Multilingual
| SWE Multilingual | 81.0 | #9 of 46 |
Instruction following
| IFBench | 81.3 | #6 of 42 |
Pricing by source
USD per 1M tokens.
| Source | Input | Output | Cached in |
|---|---|---|---|
| List price | — | — | — |
Qwen publishes Qwen3.8-Flash-Next under the Qwen Community 1.0 license for self-hosting and does not publish a distinct first-party hosted token rate for this exact experimental checkpoint. BenchLM represents the open-weight row as self-host/free-per-token before infrastructure costs. Qwen Cloud's production Qwen3.8-Flash is a separate model based on this architecture, so this row does not inherit its price or default 1M context.
Arena ratings
Human-preference ratings from LMArena. Best-rated variant shown.
| WebDev (coding) | 1629 | #21 · 8.4K votes |
Open weights
From the Hugging Face repository.
- Parameters
- 180B
- Licence
- other
- Downloads (30 days)
- 2.5M
- Repository
- Qwen/Qwen3.8-Flash-Next ↗
Speed
Measured by us through OpenRouter (median of trials).
Not available on OpenRouter, so we cannot measure it yet.
Related models
Same family
- Qwen3.8-27B59.6
- Qwen3.8 Max69.7
- Qwen3.7 Flash47.0
- Qwen3.5 Flash42.8
Similar score
- Gemini 3.6 Flash62.8 · $1.50
- Muse Spark Open62.8 · $2.00
- Qwen3.7 Max62.6 · $2.21
- DeepSeek V4 Pro 081363.2 · $0.99
- Grok 4.563.3 · $3.00