Qwen3.8-Flash-Next leads overall
Ahead on 9 of 9 benchmarks, averaging 4.2 points higher
Summarised from the 9 benchmarks both models were scored on; details in the charts below.

Qwen3.8-27B
阿里巴巴
Updates live with the mode filters below.
Best overall
Qwen3.8-Flash-Next · 66.60
Best single
Qwen3.8-Flash-Next · MathVision 95.70
Modality coverage
Qwen3.8-27B · 3 modalities
Head to head
9
Benchmarks
0
Wins
9
Losses
-4.19
Average diff
Benchmark-by-benchmark comparison. Changing the thinking mode or tool filters updates the chart and table below.
Data sourced primarily from official releases (GitHub, Hugging Face, papers), then benchmark leaderboards, then third-party evaluators. Learn about our data methodology
Each axis is the mean percentage score of one benchmark domain. It is an average, not a capability rating.
Relative edge: none clear / Relative gap: 编程与软件工程 -6.2
Relative edge: 编程与软件工程 +6.2 / Relative gap: none clear
Method: for each model and benchmark, all scores in the current mode scope are averaged (not the best score), then those benchmark scores are averaged within each domain. Only benchmarks scored on a 0-100 scale by at least two of the selected models count — Elo and rating-scale benchmarks such as Codeforces or Arena are excluded, because averaging a 1500 rating with an 85% accuracy produces a meaningless number. Missing values are not counted as zero, and the averages are unweighted, so domains with harder benchmarks read lower.
Every model and runtime mode, benchmark by benchmark. Values are comparable along a row, not between different benchmarks.
9 benchmarks with comparable scores. Each model shows its best score; mode label is displayed below.
| Benchmark | Qwen3.8-27B | Qwen3.8-Flash-Next |
|---|---|---|
HLE 综合评估 | 30.80Thinking Enabled | 35.90Thinking Level · Extra High |
GPQA Diamond 科学与综合推理 | 89.20Thinking Enabled | 91.70Thinking Level · Extra High |
DeepSWE 编程与软件工程 | 42.20Thinking Enabled | Tools | 58.70Thinking Level · Extra High | Tools |
LiveCodeBench 编程与软件工程 | 90.30Thinking Enabled | 91.90Thinking Level · Extra High |
NL2Repo-Bench 编程与软件工程 | 42.30Thinking Enabled | Tools | 48.10Thinking Level · Extra High | Tools |
SWE-Bench Pro - Public 编程与软件工程 | 61.70Thinking Enabled | Tools | 62.50Thinking Level · Extra High | Tools |
Agents' Last Exam Agent能力评测 | 20.40Thinking Enabled | Tools | 24.30Thinking Level · Extra High | Tools |
CharXiv RQ 多模态理解 | 90.20Thinking Enabled | Tools | 90.60Thinking Level · Extra High | Tools |
MathVision 多模态理解 | 94.60Thinking Enabled | Tools | 95.70Thinking Level · Extra High | Tools |
Official list prices per model API, split by input and output. Unit: USD per 1M tokens.
Architecture, licensing and API modalities. "Not provided" means the field is missing from our database.
| Features & specs | Qwen3.8-27B阿里巴巴 | Qwen3.8-Flash-Next阿里巴巴 |
|---|---|---|
Core specsRelease | 2026-08-14 | 2026-08-26 |
Context length | 256K | 256K(原生,可扩展至 1M) |
Total parameters | 27B | 125B |
Active parameters | N/A | 6B |
Max output length | 131,072 tokens | 131,072 tokens |
Architecture | Dense | MoE (mixture of experts) |
Runtime modes | 开启关闭 | 极高关闭低中 |
LicenseCode Open Source | Open Source · Apache 2.0 | Open Source · Qwen Community License 1.0 |
Weights Open Source | Open Source · Apache 2.0 | Open Source · Qwen Community License 1.0 |
Licensing status | 免费商用授权 | 有条件免费商用授权 |
Local deploymentWeight size | 约 55.6 GB(BF16,18 个 safetensors 分片) | 约 360 GB(BF16,131 个 safetensors 分片) |
VRAM for weights | ≈ 55.6 GB (weights only, excludes KV cache) | ≈ 360 GB (weights only, excludes KV cache) |
Weights | Hugging Face | Hugging Face |
Source repo | Not provided | GitHub |
API modality supportText Input/Output | / | / |
Image Input/Output | / | / |
Video Input/Output | / | / |
ResourcesPaper / report | Not provided | On the Design of Qwen3.8-Next Architecture: Evaluation, Efficiency, and Training Stability |

Qwen3.8-Flash-Next
阿里巴巴