Model comparison
GPT-6 Astra vs Qwen3.5 397B-A17B
GPT-6 Astra is the stronger model overall, scoring 70.8 to 46.0 on the Noometry Index. Qwen3.5 397B-A17B costs 15× less per token, which makes it the better buy when GPT-6 Astra's lead doesn't matter for your workload.
Last verified . 30 shared benchmarks.
Alibaba (Qwen)Summary
- They share 30 benchmarks with published results for both. GPT-6 Astra scores higher in 9 categories and Qwen3.5 397B-A17B in 1 category; 8 gaps are clear of the uncertainty.
- The widest gap is in reasoning, where GPT-6 Astra leads 85.1 to 34.5.
- The biggest single-benchmark swing is Mystery Game Puzzles: 84% for GPT-6 Astra and 18% for Qwen3.5 397B-A17B.
- Qwen3.5 397B-A17B is cheaper at $0.60 / $3.60 per million input/output tokens, against $10 / $50 for GPT-6 Astra.
- GPT-6 Astra accepts more context: 1.05M tokens versus 262K.
- Qwen3.5 397B-A17B has downloadable open weights; the other is API-only.
Side by side
| GPT-6 Astra | Qwen3.5 397B-A17B | |
|---|---|---|
| Provider | OpenAI | Alibaba (Qwen) |
| Noometry Index | 70.8 | 46.0 |
| Released | 2026-09-03 | 2026-02-01 |
| Weights | Proprietary | Open |
| Context window | 1.05M | 262K |
| Max output | 128K | 66K |
| Input $ / M tokens | $10 | $0.60 |
| Output $ / M tokens | $50 | $3.60 |
| Results tracked | 56 | 36 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding GPT-6 Astra leads
GPT-6 Astra: 73.7 (#2), Qwen3.5 397B-A17B: 42.0 (#114)
| Benchmark | GPT-6 Astra | Qwen3.5 397B-A17B |
|---|---|---|
| LMArena WebDev | 1786 | 1400 |
| LMArena Coding | 1487 | 1465 |
| DeepSWE | 74.1% | — |
| FrontierCode | 53.3% | — |
| FrontierSWE | 65.5% | — |
| SciCode | 56.5% | — |
| GSO | 79.4% | — |
| WeirdML | 93.6% | — |
| MirrorCode | 46.7% | — |
| ALE-Bench | 2,951 | — |
Agentic & Tool Use GPT-6 Astra leads
GPT-6 Astra: 52.9 (#3), Qwen3.5 397B-A17B: 33.3 (#53)
| Benchmark | GPT-6 Astra | Qwen3.5 397B-A17B |
|---|---|---|
| APEX-Agents | 64.7% | 24.9% |
| Remote Labor Index | 20.8% | — |
| τ²-bench Airline | — | 81.5% |
| τ²-bench Banking | — | 9.8% |
| τ²-bench Retail | — | 84.4% |
| τ²-bench Telecom | — | 97.8% |
| BALROG | 68.3% | — |
| GDP.pdf | 34.2% | — |
| Vending-Bench 2 | 15,515 | — |
Reasoning GPT-6 Astra leads
GPT-6 Astra: 85.1 (#1), Qwen3.5 397B-A17B: 34.5 (#70)
| Benchmark | GPT-6 Astra | Qwen3.5 397B-A17B |
|---|---|---|
| NYT Connections (extended) | 98.1% | 58.9% |
| Chess Puzzles | 72% | 13% |
| LMArena Hard Prompts | 1462 | 1448 |
| Mystery Game Puzzles | 84% | 18% |
| DTBench | 97.3% | 87.5% |
| LMCA | 64.4% | 37.9% |
| Epoch Capabilities Index | 166.45 | 146.65 |
| ARC-AGI-2 | 95% | — |
| Kagi LLM Benchmark | — | 73.7% |
| ARC-AGI-1 | 98.5% | — |
| CritPt | 31.7% | — |
| Thematic Generalization | — | 65.1% |
| EBR-Bench | 76.2% | — |
| Bench to the Future 3 | 0.14 | — |
Math GPT-6 Astra leads
GPT-6 Astra: 93.5 (#2), Qwen3.5 397B-A17B: 46.1 (#73)
| Benchmark | GPT-6 Astra | Qwen3.5 397B-A17B |
|---|---|---|
| FrontierMath (Tiers 1-3) | 93.7% | 31.2% |
| OTIS Mock AIME 2024-2025 | 100% | 88.9% |
| LMArena Math | 1465 | 1454 |
| FrontierMath Tier 4 | 97.6% | — |
| ProofBench | 99% | — |
| FrontierMath Erdős | 2.9% | — |
Knowledge GPT-6 Astra leads
GPT-6 Astra: 75.3 (#1), Qwen3.5 397B-A17B: 53.3 (#58)
| Benchmark | GPT-6 Astra | Qwen3.5 397B-A17B |
|---|---|---|
| GPQA Diamond | 95.8% | 86.4% |
| LMArena Expert | 1483 | 1462 |
| Humanity's Last Exam | 54.8% | — |
| SimpleQA Verified | 75.6% | — |
| Vectara Hallucination Rate | 8.7% | — |
Multimodal GPT-6 Astra leads
GPT-6 Astra: 55.0 (#3), Qwen3.5 397B-A17B: 40.7 (#44)
| Benchmark | GPT-6 Astra | Qwen3.5 397B-A17B |
|---|---|---|
| LMArena Vision | 1281 | 1263 |
| Blueprint-Bench 2 | 49.7% | — |
| Furniture Assembly | 80% | — |
| LMArena Document | 1468 | — |
Multilingual Too close to call
GPT-6 Astra: 53.7 (#61), Qwen3.5 397B-A17B: 53.7 (#59)
| Benchmark | GPT-6 Astra | Qwen3.5 397B-A17B |
|---|---|---|
| LMArena Non-English | 1430 | 1430 |
| LMArena Chinese | 1484 | 1500 |
| LMArena French | 1456 | 1461 |
| LMArena German | 1440 | 1447 |
| LMArena Japanese | 1379 | 1426 |
| LMArena Korean | 1426 | 1384 |
| LMArena Russian | 1436 | 1429 |
| LMArena Spanish | 1407 | 1441 |
Instruction Following GPT-6 Astra leads
GPT-6 Astra: 76.3 (#44), Qwen3.5 397B-A17B: 75.0 (#77)
| Benchmark | GPT-6 Astra | Qwen3.5 397B-A17B |
|---|---|---|
| LMArena Instruction Following | 1450 | 1424 |
Long Context Too close to call
GPT-6 Astra: 44.5 (#62), Qwen3.5 397B-A17B: 44.1 (#74)
| Benchmark | GPT-6 Astra | Qwen3.5 397B-A17B |
|---|---|---|
| LMArena Longer Query | 1456 | 1442 |
Writing & Preference GPT-6 Astra leads
GPT-6 Astra: 75.3 (#7), Qwen3.5 397B-A17B: 62.3 (#79)
| Benchmark | GPT-6 Astra | Qwen3.5 397B-A17B |
|---|---|---|
| LMArena Text | 1441 | 1438 |
| LMArena Creative Writing | 1418 | 1401 |
| EQ-Bench Creative Writing | 2173 | 1478 |
| LMArena Multi-Turn | 1448 | 1446 |
Frequently asked questions
Is GPT-6 Astra better than Qwen3.5 397B-A17B?
GPT-6 Astra is the stronger model overall, scoring 70.8 to 46.0 on the Noometry Index. Qwen3.5 397B-A17B costs 15× less per token, which makes it the better buy when GPT-6 Astra's lead doesn't matter for your workload.
Which is cheaper, GPT-6 Astra or Qwen3.5 397B-A17B?
Qwen3.5 397B-A17B is cheaper. It lists at $0.60 per million input tokens and $3.60 per million output tokens; GPT-6 Astra lists at $10 and $50.
Is GPT-6 Astra or Qwen3.5 397B-A17B better for coding?
GPT-6 Astra scores higher on coding benchmarks: 73.7 versus 42.0 in the Noometry coding category.
Which has the bigger context window?
GPT-6 Astra does, with 1.05M tokens against 262K.
How many benchmarks do GPT-6 Astra and Qwen3.5 397B-A17B share?
30 benchmarks have published results for both models. GPT-6 Astra has 56 scored results on Noometry and Qwen3.5 397B-A17B has 36.