# GPT-5.4 vs Qwen3.5 397B

> Side-by-side benchmark comparison for GPT-5.4 and Qwen3.5 397B across agentic, coding, multimodal, knowledge, reasoning, multilingual, and math tasks.

- Canonical page: https://benchlm.ai/compare/gpt-5-4-vs-qwen3-5-397b
- Last updated: October 17, 2026

- Shared sourced benchmarks: 17
- HTML indexing: indexable
- Ranking lane: BenchAlign v5.8

## Quick Verdict

GPT-5.4 leads by point estimate. Conditional score ranges do not establish rank confidence. Choose using the category evidence and your own workload tests.

## Summary

- GPT-5.4 has the higher overall point estimate 66.7 to 55.34. Conditional score ranges do not establish rank confidence.
- The clearest category separation is in knowledge, where the averages are 69.1 for GPT-5.4 and 53.1 for Qwen3.5 397B.
- The biggest single benchmark swing is MCP Atlas in Agentic, with scores of 70.6% and 46.1%.
- Qwen3.5 397B is the cheaper option on output tokens, which matters if you expect large responses or heavy interactive use.
- GPT-5.4 also has the larger context window at 1.05M.

## Model Snapshot

| Property | GPT-5.4 | Qwen3.5 397B |
|----------|----------|----------|
| Creator | OpenAI | Alibaba |
| Type | Proprietary | Open Weight |
| Reasoning | Reasoning | Non-Reasoning |
| Context | 1.05M | 128K |
| Overall Score | 66.7 | 55.34 |
| Benchmarks Covered | 42 | 62 |
| Pricing (input/output) | $2.50 / $15.00 | $0.60 / $3.60 |

## Category Breakdown

### Agentic

- Winner: Directional only
- GPT-5.4 public-lane score: 52.1 (Supported · #44/128)
- Qwen3.5 397B public-lane score: 43.8 (Estimated · #61/128)

| Benchmark | GPT-5.4 | Qwen3.5 397B | Winner |
|-----------|-----------|-----------|--------|
| Terminal-Bench 2.0 | 75.1% | 52.5% | GPT-5.4 |
| CyberGym | 79.0% | Coming soon | Coming soon |
| BrowseComp | 82.7% | 69% | GPT-5.4 |
| OSWorld-Verified | 75% | 62.2% | GPT-5.4 |
| MCP Atlas | 70.6% | 46.1% | GPT-5.4 |
| Toolathlon | 54.6% | 36.3% | GPT-5.4 |
| τ²-bench results | 98.9% | Coming soon | Coming soon |
| Claw-Eval | 60.3% | 56.8% | GPT-5.4 |
| DeepSearchQA | 73.6% | Coming soon | Coming soon |
| Gert Labs | 64.89% | 46.76% | GPT-5.4 |
| ResearchClawBench | 15.3% | 14.2% | GPT-5.4 |
| JobBench | 38.9% | Coming soon | Coming soon |
| ExploitGym | 6.0% | Coming soon | Coming soon |
| ApprenticeBench | 11% | Coming soon | Coming soon |
| MazeBench | 0% | Coming soon | Coming soon |
| QwenClawBench | Coming soon | 51.8% | Coming soon |
| τ³-bench results | Coming soon | 68.4% | Coming soon |
| VITA-Bench | Coming soon | 43.7% | Coming soon |
| DeepPlanning | Coming soon | 37.6% | Coming soon |
| MCP-Tasks | Coming soon | 74.2% | Coming soon |
| WideResearch | Coming soon | 74.0% | Coming soon |

### Coding

- Winner: Not comparable
- GPT-5.4 public-lane score: 47.7 (Supported · #48/150)
- Qwen3.5 397B public-lane score: Coming soon

| Benchmark | GPT-5.4 | Qwen3.5 397B | Winner |
|-----------|-----------|-----------|--------|
| LiveCodeBench Pro | 87.5% | Coming soon | Coming soon |
| SWE-bench Pro | 57.7% | 50.9% | GPT-5.4 |
| React Native Evals | 85.3% | Coming soon | Coming soon |
| Terminal-Bench 2.0 | 75.1% | 52.5% | GPT-5.4 |
| Vibe Code Bench | 67.42% | Coming soon | Coming soon |
| PostTrainBench v1.1 | 19.0% | Coming soon | Coming soon |
| SWE-bench Verified | Coming soon | 76.2% | Coming soon |
| LiveCodeBench v6 | Coming soon | 83.6% | Coming soon |
| SWE Multilingual | Coming soon | 69.3% | Coming soon |
| NL2Repo | Coming soon | 32.2% | Coming soon |

### Reasoning

- Winner: Not comparable
- GPT-5.4 public-lane score: 64.2 (#23/28)
- Qwen3.5 397B public-lane score: 74.9 (Unranked · 2 rankable rows)

| Benchmark | GPT-5.4 | Qwen3.5 397B | Winner |
|-----------|-----------|-----------|--------|
| ARC-AGI-2 | 74.0% | Coming soon | Coming soon |
| ARC-AGI-3 | 0.2% | Coming soon | Coming soon |
| ARC-AGI-1 | 93.67% | Coming soon | Coming soon |
| LongBench v2 | Coming soon | 63.2% | Coming soon |
| AI-Needle | Coming soon | 68.7% | Coming soon |

### Multimodal & Grounded

- Winner: Directional only
- GPT-5.4 public-lane score: 73.1 (#30/54)
- Qwen3.5 397B public-lane score: 73.6 (#29/54)

| Benchmark | GPT-5.4 | Qwen3.5 397B | Winner |
|-----------|-----------|-----------|--------|
| MMMU-Pro | 81.2% | 79% | GPT-5.4 |
| OfficeQA Pro | 53.2% | Coming soon | Coming soon |
| MMMU-Pro w/ Python | 82.1% | Coming soon | Coming soon |
| CharXiv | 82.8% | 80.8% | GPT-5.4 |
| ERQA | 65.4% | 67.5% | Qwen3.5 397B |
| SimpleVQA | 61.1% | 67.1% | Qwen3.5 397B |
| ScreenSpot Pro | 85.4% | 65.6% | GPT-5.4 |
| ZeroBench | 41.0% | Coming soon | Coming soon |
| MedXpertQA (MM) | 77.1% | Coming soon | Coming soon |
| MMMU | Coming soon | 85.0% | Coming soon |
| RealWorldQA | Coming soon | 83.9% | Coming soon |
| OmniDocBench 1.5 | Coming soon | 90.8% | Coming soon |
| Video-MME (with subtitle) | Coming soon | 87.5% | Coming soon |
| Video-MME (w/o subtitle) | Coming soon | 83.7% | Coming soon |
| MathVision | Coming soon | 88.6% | Coming soon |
| We-Math | Coming soon | 87.9% | Coming soon |
| DynaMath | Coming soon | 86.3% | Coming soon |
| MMLongBench-Doc | Coming soon | 61.5% | Coming soon |
| CC-OCR | Coming soon | 82.0% | Coming soon |
| AI2D_TEST | Coming soon | 93.9% | Coming soon |
| CountBench | Coming soon | 97.2% | Coming soon |
| RefCOCO (avg) | Coming soon | 92.3% | Coming soon |
| ODINW13 | Coming soon | 47.0% | Coming soon |
| VideoMMMU | Coming soon | 84.7% | Coming soon |
| MLVU (M-Avg) | Coming soon | 86.7% | Coming soon |
| V* | Coming soon | 95.8% | Coming soon |

### Knowledge

- Winner: Directional only
- GPT-5.4 public-lane score: 69.1 (Supported · #20/184)
- Qwen3.5 397B public-lane score: 53.1 (Estimated · #68/184)

| Benchmark | GPT-5.4 | Qwen3.5 397B | Winner |
|-----------|-----------|-----------|--------|
| GPQA | 92.8% | 88.4% | GPT-5.4 |
| HLE | 52.1% | 28.7% | GPT-5.4 |
| HLE w/o tools | 39.8% | Coming soon | Coming soon |
| GPQA-D | 92.8% | Coming soon | Coming soon |
| HealthBench Hard | 40.1% | Coming soon | Coming soon |
| MedXpertQA (Text) | 59.6% | Coming soon | Coming soon |
| HealthBench Professional | 48.1% | Coming soon | Coming soon |
| SuperGPQA | Coming soon | 70.4% | Coming soon |
| MMLU-Pro | Coming soon | 87.8% | Coming soon |
| MMLU-Redux | Coming soon | 94.9% | Coming soon |
| C-Eval | Coming soon | 93% | Coming soon |

### Multilingual

- Winner: Not comparable
- GPT-5.4 public-lane score: Coming soon
- Qwen3.5 397B public-lane score: 93.8 (#5/17)

| Benchmark | GPT-5.4 | Qwen3.5 397B | Winner |
|-----------|-----------|-----------|--------|
| MMLU-ProX | Coming soon | 84.7% | Coming soon |
| NOVA-63 | Coming soon | 59.1% | Coming soon |
| INCLUDE | Coming soon | 85.6% | Coming soon |
| PolyMath | Coming soon | 73.3% | Coming soon |
| MAXIFE | Coming soon | 88.2% | Coming soon |

### Instruction Following

- Winner: Directional only
- GPT-5.4 public-lane score: 90.7 (#23/148)
- Qwen3.5 397B public-lane score: 79.9 (#59/148)

| Benchmark | GPT-5.4 | Qwen3.5 397B | Winner |
|-----------|-----------|-----------|--------|
| IFEval | Coming soon | 92.6% | Coming soon |
| IFBench | Coming soon | 76.5% | Coming soon |

### Mathematics

- Winner: Not comparable
- GPT-5.4 public-lane score: 82 (Unranked · 2 rankable rows)
- Qwen3.5 397B public-lane score: 81.7 (Unranked · 5 rankable rows)

| Benchmark | GPT-5.4 | Qwen3.5 397B | Winner |
|-----------|-----------|-----------|--------|
| FrontierMath v2 (Tiers 1-3) | 78.596% | Coming soon | Coming soon |
| FrontierMath v2 (Tier 4) | 49.000% | Coming soon | Coming soon |
| AIME26 | Coming soon | 93.3% | Coming soon |
| HMMT Feb 2025 | Coming soon | 94.8% | Coming soon |
| HMMT Nov 2025 | Coming soon | 92.7% | Coming soon |
| HMMT Feb 2026 | Coming soon | 87.9% | Coming soon |
| MMAnswerBench | Coming soon | 80.9% | Coming soon |

## FAQ

### Which is better overall, GPT-5.4 or Qwen3.5 397B?

GPT-5.4 has the higher overall point estimate. Conditional score ranges do not establish rank confidence.

### Where is the biggest gap between GPT-5.4 and Qwen3.5 397B?

The widest category gap is in knowledge, where the averages are 69.1 for GPT-5.4 and 53.1 for Qwen3.5 397B.

### How many benchmarks does GPT-5.4 cover on BenchLM?

GPT-5.4 currently has 42 sourced benchmark scores on BenchLM.

### How many benchmarks does Qwen3.5 397B cover on BenchLM?

Qwen3.5 397B currently has 62 sourced benchmark scores on BenchLM.

## Related Comparisons

- [GPT-5.4 vs GPT-5.4 Pro](/compare/gpt-5-4-vs-gpt-5-4-pro)
- [Qwen3.5 397B vs GPT-5.4 Pro](/compare/gpt-5-4-pro-vs-qwen3-5-397b)
- [GPT-5.4 vs GPT-5.4 mini](/compare/gpt-5-4-vs-gpt-5-4-mini)
- [Qwen3.5 397B vs GPT-5.4 mini](/compare/gpt-5-4-mini-vs-qwen3-5-397b)

## Explore More

- [GPT-5.4 profile](/models/gpt-5-4)
- [Qwen3.5 397B profile](/models/qwen3-5-397b)
- [Compare Pricing](/llm-pricing)
- [Alternative Finder](/tools/alternative-finder)
- [LLM Selector](/tools/llm-selector)
- [Overall Rankings](/best/overall)