Model comparison
Qwen3.5 397B vs Qwen3.6-27B
Head-to-head evidence from 37 shared benchmark results across 7 categories. Overall scores shown here use the public BenchAlign v5 ranking lane.
Public leaderboard positions: Qwen3.5 397B #71 (Estimated); Qwen3.6-27B #93 (Estimated). Intervals and evidence labels describe ranking uncertainty, not a guarantee for a specific workload.
Evidence parity. Qwen3.5 397B and Qwen3.6-27B share 37 comparable benchmark results. 5 of 8 categories are comparable. 18 results are unique to Qwen3.5 397B; 17 to Qwen3.6-27B.
Updated July 21, 2026- Shared results
- 37
- Qwen3.5 397B only
- 18
- Qwen3.6-27B only
- 17
- Comparable categories
- 5 / 8
Pick Qwen3.5 397B if you want the stronger benchmark profile. Qwen3.6-27B only becomes the better choice if coding is the priority or you want the cheaper token bill.
Confidence note. This is a partial-evidence comparison with 37 shared benchmark results across 7 evidence categories; 5 of 8 categories currently have scoreable aggregates for both models. Treat the verdict as directional until coverage is more balanced.
Why this result
Qwen3.5 397B is clearly ahead on the BenchAlign aggregate, 57.01 to 53.82. The gap is large enough that you do not need to squint at the spreadsheet to see the difference.
Qwen3.5 397B's sharpest advantage is in knowledge, where it averages 56.6 against 53.3. The single biggest benchmark swing on the page is Terminal-Bench 2.0, 52.5% to 59.3%. Qwen3.6-27B does hit back in coding, so the answer changes if that is the part of the workload you care about most.
Qwen3.5 397B is also the more expensive model on tokens at $0.60 input / $3.60 output per 1M tokens, versus $0.00 input / $0.00 output per 1M tokens for Qwen3.6-27B. That is roughly Infinityx on output cost alone. Qwen3.6-27B is the reasoning model in the pair, while Qwen3.5 397B is not. That usually helps on harder chain-of-thought-heavy tests, but it can also mean more latency and more token spend in real use. Qwen3.6-27B gives you the larger context window at 262K, compared with 128K for Qwen3.5 397B.
Category breakdown
Exact category averages are shown below. Not measured means BenchLM does not have enough sourced public coverage for that model and category.
| Category | Qwen3.5 397B | Δ | Qwen3.6-27B |
|---|---|---|---|
| Coding | Qwen3.5 397B66.5 | Margin→ 11.0 | Qwen3.6-27B77.5 |
| Knowledge | Qwen3.5 397B56.6 | Margin← 3.3 | Qwen3.6-27B53.3 |
| Multimodal | Qwen3.5 397B79.6 | Margin← 2.9 | Qwen3.6-27B76.7 |
| Agentic | Qwen3.5 397B56.5 | Margin→ 2.8 | Qwen3.6-27B59.3 |
| Math | Qwen3.5 397B90.6 | Margin← 1.4 | Qwen3.6-27B89.2 |
| Reasoning | Qwen3.5 397B63.2 | MarginNo overlap | Qwen3.6-27BNot measured |
| Multilingual | Qwen3.5 397B84.7 | MarginNo overlap | Qwen3.6-27BNot measured |
| Inst. Following | Qwen3.5 397B92.6 | MarginNo overlap | Qwen3.6-27BNot measured |
Decisive benchmark drivers
The largest measured benchmark gaps in this matchup, with exact reported values.
More
- Source ↗
Terminal-Bench 2.0
AgenticA 52.5%B 59.3%Winner: Qwen3.6-27BΔ 6.8Terminal-Bench 2.0: Qwen3.5 397B scored 52.5%; Qwen3.6-27B scored 59.3%. Qwen3.6-27B wins this benchmark. - Source ↗
HLE
KnowledgeA 28.7%B 24%Winner: Qwen3.5 397BΔ 4.7HLE: Qwen3.5 397B scored 28.7%; Qwen3.6-27B scored 24%. Qwen3.5 397B wins this benchmark. - Source ↗
SuperGPQA
KnowledgeA 70.4%B 66%Winner: Qwen3.5 397BΔ 4.4SuperGPQA: Qwen3.5 397B scored 70.4%; Qwen3.6-27B scored 66%. Qwen3.5 397B wins this benchmark. - Source ↗
HMMT Feb 2026
MathA 87.9%B 84.3%Winner: Qwen3.5 397BΔ 3.6HMMT Feb 2026: Qwen3.5 397B scored 87.9%; Qwen3.6-27B scored 84.3%. Qwen3.5 397B wins this benchmark. - Source ↗
MMMU-Pro
MultimodalA 79%B 75.8%Winner: Qwen3.5 397BΔ 3.2MMMU-Pro: Qwen3.5 397B scored 79%; Qwen3.6-27B scored 75.8%. Qwen3.5 397B wins this benchmark.
Operational comparison
Runtime and commercial metrics are compared only when both models have a complete sourced value.
| Metric | Qwen3.5 397B | Qwen3.6-27B | Comparison |
|---|---|---|---|
| Input / output priceUSD per 1M tokens | Qwen3.5 397B$0.6 input / $3.6 output | Qwen3.6-27B$0 input / $0 output | Qwen3.6-27B has the lower combined listed price. |
| Generation speedtokens per second | Qwen3.5 397B96 tok/s | Qwen3.6-27BNot available | A complete speed comparison is not available. |
| First-answer latencyseconds to first token | Qwen3.5 397B2.44 s | Qwen3.6-27BNot available | A complete latency comparison is not available. |
| Context windowmaximum listed tokens | Qwen3.5 397B128K | Qwen3.6-27B262K | Qwen3.6-27B lists the larger context window. |
Benchmark Deep Dive
AgenticQwen3.6-27B wins20 benchmarks
| Benchmark | Qwen3.5 397B | Qwen3.6-27B | Result |
|---|---|---|---|
| Terminal-Bench 2.0Source | 52.5% | 59.3% | Qwen3.6-27B leads |
| BrowseCompSource | 62% | — | Not comparable |
| Claw-EvalSource | 56.8% | 72.4% | Qwen3.6-27B leads |
| QwenClawBenchSource | 51.8% | 53.4% | Qwen3.6-27B leads |
| τ³-bench resultsSource | 68.4% | — | Not comparable |
| VITA-BenchSource | 43.7% | — | Not comparable |
| DeepPlanningSource | 37.6% | — | Not comparable |
| ToolathlonSource | 36.3% | — | Not comparable |
| MCP AtlasSource | 46.1% | — | Not comparable |
| MCP-TasksSource | 74.2% | — | Not comparable |
| WideResearchSource | 74.0% | — | Not comparable |
| τ²-bench resultsSource | 95.6% | 94.2% | Qwen3.5 397B leads |
| Gert LabsSource | 46.76% | 54.84% | Qwen3.6-27B leads |
| ResearchClawBenchSource | 14.2% | — | Not comparable |
| AA Agentic IndexSource | 19.9% | 27.0% | Qwen3.6-27B leads |
| APEX-Agents-AASource | 15.3% | — | Not comparable |
| GDPval-AASource | 23.1% | 32.0% | Qwen3.6-27B leads |
| GDPval-AASource | 962 | 1140 | Qwen3.6-27B leads |
| QwenWebBenchSource | — | 1487 | Not comparable |
| AndroidWorldSource | — | 70.3% | Not comparable |
CodingQwen3.6-27B wins9 benchmarks
| Benchmark | Qwen3.5 397B | Qwen3.6-27B | Result |
|---|---|---|---|
| SWE-bench VerifiedSource | 76.2% | 77.2% | Qwen3.6-27B leads |
| LiveCodeBench v6Source | 83.6% | — | Not comparable |
| SWE-bench ProSource | 50.9% | 53.5% | Qwen3.6-27B leads |
| AA-SciCodeSource | 42.0% | 39.8% | Qwen3.5 397B leads |
| AA Coding IndexSource | 48.2% | 53.7% | Qwen3.6-27B leads |
| SWE MultilingualSource | — | 71.3% | Not comparable |
| Terminal-Bench 2.0Source | — | 59.3% | Not comparable |
| LiveCodeBenchSource | — | 83.9% | Not comparable |
| NL2RepoSource | — | 36.2% | Not comparable |
Reasoning4 benchmarks
KnowledgeQwen3.5 397B wins12 benchmarks
| Benchmark | Qwen3.5 397B | Qwen3.6-27B | Result |
|---|---|---|---|
| GPQASource | 88.4% | 87.8% | Qwen3.5 397B leads |
| SuperGPQASource | 70.4% | 66% | Qwen3.5 397B leads |
| MMLU-ProSource | 87.8% | 86.2% | Qwen3.5 397B leads |
| MMLU-ReduxSource | 94.9% | 93.5% | Qwen3.5 397B leads |
| C-EvalSource | 93% | 91.4% | Qwen3.5 397B leads |
| HLESource | 28.7% | 24% | Qwen3.5 397B leads |
| Artificial Analysis Intelligence IndexSource | 33.7% | 37.0% | Qwen3.6-27B leads |
| AA-GPQA DiamondSource | 89.3% | 84.2% | Qwen3.5 397B leads |
| AA-HLESource | 27.3% | 21.6% | Qwen3.5 397B leads |
| AA-Omniscience IndexSource | -29.8% | -19.8% | Qwen3.6-27B leads |
| AA-Omniscience AccuracySource | 31.4% | 19.2% | Qwen3.5 397B leads |
| AA-Omniscience Hallucination RateSource | 89.1% | 48.3% | Qwen3.6-27B leads |
MathQwen3.5 397B wins5 benchmarks
Multilingual2 benchmarks
MultimodalQwen3.5 397B wins18 benchmarks
| Benchmark | Qwen3.5 397B | Qwen3.6-27B | Result |
|---|---|---|---|
| MMMU-ProSource | 79% | 75.8% | Qwen3.5 397B leads |
| MathVisionSource | 88.6% | — | Not comparable |
| CharXivSource | 80.8% | 78.4% | Qwen3.5 397B leads |
| VideoMMMUSource | 84.7% | 84.4% | Qwen3.5 397B leads |
| ScreenSpot ProSource | 65.6% | — | Not comparable |
| V*Source | 95.8% | 94.7% | Qwen3.5 397B leads |
| AA-MMMU-ProSource | 77.3% | 74.6% | Qwen3.5 397B leads |
| MMMUSource | — | 82.9% | Not comparable |
| RealWorldQASource | — | 84.1% | Not comparable |
| DynaMathSource | — | 85.6% | Not comparable |
| MStarSource | — | 81.4% | Not comparable |
| SimpleVQASource | — | 56.1% | Not comparable |
| CC-OCRSource | — | 81.2% | Not comparable |
| CountBenchSource | — | 97.8% | Not comparable |
| RefCOCO (avg)Source | — | 92.5% | Not comparable |
| ERQASource | — | 62.5% | Not comparable |
| Video-MME (with subtitle)Source | — | 87.7% | Not comparable |
| MLVU (M-Avg)Source | — | 86.6% | Not comparable |
Frequently Asked Questions (6)
Which is better, Qwen3.5 397B or Qwen3.6-27B?
Qwen3.5 397B is ahead on BenchLM's BenchAlign leaderboard, 57.01 to 53.82. The biggest single separator in this matchup is Terminal-Bench 2.0, where the scores are 52.5% and 59.3%.
Which is better for knowledge tasks, Qwen3.5 397B or Qwen3.6-27B?
Qwen3.5 397B has the edge for knowledge tasks in this comparison, averaging 56.6 versus 53.3. Inside this category, AA-Omniscience Hallucination Rate is the benchmark that creates the most daylight between them.
Which is better for coding, Qwen3.5 397B or Qwen3.6-27B?
Qwen3.6-27B has the edge for coding in this comparison, averaging 77.5 versus 66.5. Inside this category, AA Coding Index is the benchmark that creates the most daylight between them.
Which is better for math, Qwen3.5 397B or Qwen3.6-27B?
Qwen3.5 397B has the edge for math in this comparison, averaging 90.6 versus 89.2. Inside this category, HMMT Feb 2026 is the benchmark that creates the most daylight between them.
Which is better for agentic tasks, Qwen3.5 397B or Qwen3.6-27B?
Qwen3.6-27B has the edge for agentic tasks in this comparison, averaging 59.3 versus 56.5. Inside this category, GDPval-AA is the benchmark that creates the most daylight between them.
Which is better for multimodal and grounded tasks, Qwen3.5 397B or Qwen3.6-27B?
Qwen3.5 397B has the edge for multimodal and grounded tasks in this comparison, averaging 79.6 versus 76.7. Inside this category, MMMU-Pro is the benchmark that creates the most daylight between them.
Self-host vs API cost
Estimates at 50,000 req/day · 1000 tokens/req average.
Related Comparisons
Explore More
Choose a model with this week’s evidence
Join 2,000+ readers for ranking moves, pricing changes, and the claims that still need proof.
One email each week. Unsubscribe anytime.