Model comparison
GPT-5.6 Sol vs GPT-6 Astra
GPT-6 Astra is the stronger model overall, scoring 70.8 to 65.0 on the Noometry Index. GPT-5.6 Sol costs 2.5× less per token, which makes it the better buy when GPT-6 Astra's lead doesn't matter for your workload.
Last verified . 54 shared benchmarks.
Summary
- They share 54 benchmarks with published results for both. GPT-5.6 Sol scores higher in 3 categories and GPT-6 Astra in 7 categories; 9 gaps are clear of the uncertainty.
- The widest gap is in knowledge, where GPT-6 Astra leads 75.3 to 64.3.
- The biggest single-benchmark swing is FrontierSWE: 32.2% for GPT-5.6 Sol and 65.5% for GPT-6 Astra.
- GPT-5.6 Sol is cheaper at $4 / $20 per million input/output tokens, against $10 / $50 for GPT-6 Astra.
Side by side
| GPT-5.6 Sol | GPT-6 Astra | |
|---|---|---|
| Provider | OpenAI | OpenAI |
| Noometry Index | 65.0 | 70.8 |
| Released | 2026-07-09 | 2026-09-03 |
| Weights | Proprietary | Proprietary |
| Context window | 1.05M | 1.05M |
| Max output | 128K | 128K |
| Input $ / M tokens | $4 | $10 |
| Output $ / M tokens | $20 | $50 |
| Results tracked | 65 | 56 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding GPT-6 Astra leads
GPT-5.6 Sol: 65.1 (#7), GPT-6 Astra: 73.7 (#2)
| Benchmark | GPT-5.6 Sol | GPT-6 Astra |
|---|---|---|
| DeepSWE | 72.7% | 74.1% |
| FrontierCode | 47.5% | 53.3% |
| LMArena WebDev | 1618 | 1786 |
| FrontierSWE | 32.2% | 65.5% |
| SciCode | 57.1% | 56.5% |
| GSO | 76.5% | 79.4% |
| WeirdML | 89.4% | 93.6% |
| LMArena Coding | 1498 | 1487 |
| MirrorCode | 20% | 46.7% |
| ALE-Bench | 2,177 | 2,951 |
| CursorBench | 41.7% | — |
Agentic & Tool Use GPT-6 Astra leads
GPT-5.6 Sol: 50.3 (#7), GPT-6 Astra: 52.9 (#3)
| Benchmark | GPT-5.6 Sol | GPT-6 Astra |
|---|---|---|
| APEX-Agents | 51.4% | 64.7% |
| BALROG | 60% | 68.3% |
| GDP.pdf | 30.7% | 34.2% |
| Vending-Bench 2 | 9,619 | 15,515 |
| OSWorld 2.0 | 27.3% | — |
| Remote Labor Index | — | 20.8% |
| τ²-bench Banking | 46.9% | — |
| PostTrainBench | 36.2% | — |
| GBAEval | 52.6% | — |
| LMArena Search | 1257 | — |
Reasoning GPT-6 Astra leads
GPT-5.6 Sol: 74.8 (#8), GPT-6 Astra: 85.1 (#1)
| Benchmark | GPT-5.6 Sol | GPT-6 Astra |
|---|---|---|
| ARC-AGI-2 | 92.5% | 95% |
| NYT Connections (extended) | 93.8% | 98.1% |
| ARC-AGI-1 | 97.5% | 98.5% |
| CritPt | 32.3% | 31.7% |
| Chess Puzzles | 64% | 72% |
| EBR-Bench | 44.8% | 76.2% |
| LMArena Hard Prompts | 1484 | 1462 |
| Mystery Game Puzzles | 58% | 84% |
| DTBench | 96% | 97.3% |
| LMCA | 59.2% | 64.4% |
| Bench to the Future 3 | 0.14 | 0.14 |
| Epoch Capabilities Index | 161.66 | 166.45 |
| SimpleBench | 71.7% | — |
| Kagi LLM Benchmark | 67% | — |
| EnigmaEval | 37.1% | — |
| Surface Evolver Bench | 93.1% | — |
Math GPT-6 Astra leads
GPT-5.6 Sol: 85.6 (#9), GPT-6 Astra: 93.5 (#2)
| Benchmark | GPT-5.6 Sol | GPT-6 Astra |
|---|---|---|
| FrontierMath (Tiers 1-3) | 89.1% | 93.7% |
| FrontierMath Tier 4 | 82.9% | 97.6% |
| OTIS Mock AIME 2024-2025 | 100% | 100% |
| ProofBench | 83% | 99% |
| LMArena Math | 1474 | 1465 |
| FrontierMath Erdős | 0% | 2.9% |
Knowledge GPT-6 Astra leads
GPT-5.6 Sol: 64.3 (#18), GPT-6 Astra: 75.3 (#1)
| Benchmark | GPT-5.6 Sol | GPT-6 Astra |
|---|---|---|
| GPQA Diamond | 93.5% | 95.8% |
| SimpleQA Verified | 69.7% | 75.6% |
| Vectara Hallucination Rate | 12.4% | 8.7% |
| LMArena Expert | 1516 | 1483 |
| Humanity's Last Exam | — | 54.8% |
Multimodal GPT-6 Astra leads
GPT-5.6 Sol: 48.6 (#9), GPT-6 Astra: 55.0 (#3)
| Benchmark | GPT-5.6 Sol | GPT-6 Astra |
|---|---|---|
| LMArena Vision | 1281 | 1281 |
| Blueprint-Bench 2 | 33.6% | 49.7% |
| Furniture Assembly | 56.7% | 80% |
| LMArena Document | 1483 | 1468 |
Multilingual GPT-5.6 Sol leads
GPT-5.6 Sol: 55.3 (#32), GPT-6 Astra: 53.7 (#61)
| Benchmark | GPT-5.6 Sol | GPT-6 Astra |
|---|---|---|
| LMArena Non-English | 1452 | 1430 |
| LMArena Chinese | 1527 | 1484 |
| LMArena French | 1477 | 1456 |
| LMArena German | 1476 | 1440 |
| LMArena Japanese | 1471 | 1379 |
| LMArena Korean | 1442 | 1426 |
| LMArena Russian | 1468 | 1436 |
| LMArena Spanish | 1441 | 1407 |
Instruction Following GPT-5.6 Sol leads
GPT-5.6 Sol: 77.7 (#16), GPT-6 Astra: 76.3 (#44)
| Benchmark | GPT-5.6 Sol | GPT-6 Astra |
|---|---|---|
| LMArena Instruction Following | 1482 | 1450 |
Long Context Too close to call
GPT-5.6 Sol: 45.4 (#42), GPT-6 Astra: 44.5 (#62)
| Benchmark | GPT-5.6 Sol | GPT-6 Astra |
|---|---|---|
| LMArena Longer Query | 1480 | 1456 |
Writing & Preference GPT-6 Astra leads
GPT-5.6 Sol: 73.3 (#12), GPT-6 Astra: 75.3 (#7)
| Benchmark | GPT-5.6 Sol | GPT-6 Astra |
|---|---|---|
| LMArena Text | 1457 | 1441 |
| LMArena Creative Writing | 1448 | 1418 |
| EQ-Bench Creative Writing | 1972 | 2173 |
| LMArena Multi-Turn | 1460 | 1448 |
| EQ-Bench 4 | 1250 | — |
Frequently asked questions
Is GPT-5.6 Sol better than GPT-6 Astra?
GPT-6 Astra is the stronger model overall, scoring 70.8 to 65.0 on the Noometry Index. GPT-5.6 Sol costs 2.5× 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-5.6 Sol or GPT-6 Astra?
GPT-5.6 Sol is cheaper. It lists at $4 per million input tokens and $20 per million output tokens; GPT-6 Astra lists at $10 and $50.
Is GPT-5.6 Sol or GPT-6 Astra better for coding?
GPT-6 Astra scores higher on coding benchmarks: 73.7 versus 65.1 in the Noometry coding category.
Which has the bigger context window?
Both accept 1.05M tokens.
How many benchmarks do GPT-5.6 Sol and GPT-6 Astra share?
54 benchmarks have published results for both models. GPT-5.6 Sol has 65 scored results on Noometry and GPT-6 Astra has 56.