Model comparison
GPT-5.4 vs GPT-5.6 Sol
GPT-5.6 Sol is the stronger model overall, scoring 65.0 to 59.4 on the Noometry Index.
Last verified . 55 shared benchmarks.
Summary
- They share 55 benchmarks with published results for both. GPT-5.4 scores higher in 3 categories and GPT-5.6 Sol in 7 categories; 8 gaps are clear of the uncertainty.
- The widest gap is in reasoning, where GPT-5.6 Sol leads 74.8 to 61.8.
- The biggest single-benchmark swing is GSO: 31.4% for GPT-5.4 and 76.5% for GPT-5.6 Sol.
- GPT-5.4 is cheaper at $2.50 / $15 per million input/output tokens, against $4 / $20 for GPT-5.6 Sol.
Side by side
| GPT-5.4 | GPT-5.6 Sol | |
|---|---|---|
| Provider | OpenAI | OpenAI |
| Noometry Index | 59.4 | 65.0 |
| Released | 2026-03-05 | 2026-07-09 |
| Weights | Proprietary | Proprietary |
| Context window | 1.05M | 1.05M |
| Max output | 128K | 128K |
| Input $ / M tokens | $2.50 | $4 |
| Output $ / M tokens | $15 | $20 |
| Results tracked | 68 | 65 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding GPT-5.6 Sol leads
GPT-5.4: 52.6 (#33), GPT-5.6 Sol: 65.1 (#7)
| Benchmark | GPT-5.4 | GPT-5.6 Sol |
|---|---|---|
| DeepSWE | 51.8% | 72.7% |
| LMArena WebDev | 1465 | 1618 |
| SciCode | 56.6% | 57.1% |
| GSO | 31.4% | 76.5% |
| WeirdML | 77.7% | 89.4% |
| LMArena Coding | 1497 | 1498 |
| MirrorCode | 15.6% | 20% |
| ALE-Bench | 1,607 | 2,177 |
| SWE-bench Verified | 76.9% | — |
| FrontierCode | — | 47.5% |
| CursorBench | — | 41.7% |
| FrontierSWE | — | 32.2% |
| AlgoTune | 1.85 | — |
Agentic & Tool Use GPT-5.6 Sol leads
GPT-5.4: 46.5 (#13), GPT-5.6 Sol: 50.3 (#7)
| Benchmark | GPT-5.4 | GPT-5.6 Sol |
|---|---|---|
| APEX-Agents | 52.4% | 51.4% |
| τ²-bench Banking | 39.4% | 46.9% |
| PostTrainBench | 19% | 36.2% |
| GBAEval | 45.1% | 52.6% |
| LMArena Search | 1197 | 1257 |
| Vending-Bench 2 | 6,144 | 9,619 |
| Terminal-Bench | 81.8% | — |
| OSWorld 2.0 | — | 27.3% |
| DeepResearch Bench | 35.1% | — |
| BALROG | — | 60% |
| GDP.pdf | — | 30.7% |
| METR Time Horizons | 74.3% | — |
Reasoning GPT-5.6 Sol leads
GPT-5.4: 61.8 (#19), GPT-5.6 Sol: 74.8 (#8)
| Benchmark | GPT-5.4 | GPT-5.6 Sol |
|---|---|---|
| ARC-AGI-2 | 74% | 92.5% |
| Kagi LLM Benchmark | 63.8% | 67% |
| NYT Connections (extended) | 91.3% | 93.8% |
| ARC-AGI-1 | 93.7% | 97.5% |
| CritPt | 23.4% | 32.3% |
| Chess Puzzles | 44% | 64% |
| EnigmaEval | 16% | 37.1% |
| EBR-Bench | 25.4% | 44.8% |
| LMArena Hard Prompts | 1485 | 1484 |
| Mystery Game Puzzles | 37% | 58% |
| DTBench | 94.4% | 96% |
| LMCA | 52% | 59.2% |
| Epoch Capabilities Index | 156.81 | 161.66 |
| SimpleBench | — | 71.7% |
| Thematic Generalization | 80% | — |
| Surface Evolver Bench | — | 93.1% |
| Bench to the Future 3 | — | 0.14 |
| ForecastBench | 59.5 | — |
Math GPT-5.6 Sol leads
GPT-5.4: 73.5 (#19), GPT-5.6 Sol: 85.6 (#9)
| Benchmark | GPT-5.4 | GPT-5.6 Sol |
|---|---|---|
| FrontierMath (Tiers 1-3) | 78.6% | 89.1% |
| FrontierMath Tier 4 | 49% | 82.9% |
| OTIS Mock AIME 2024-2025 | 97.8% | 100% |
| ProofBench | 56% | 83% |
| LMArena Math | 1488 | 1474 |
| MathArena Final-Answer Competitions | 83.1% | — |
| FrontierMath (Feb 2025 set) | 47.6% | — |
| FrontierMath Erdős | — | 0% |
| FrontierMath Tier 4 (v1) | 27.1% | — |
Knowledge GPT-5.4 leads
GPT-5.4: 65.3 (#14), GPT-5.6 Sol: 64.3 (#18)
| Benchmark | GPT-5.4 | GPT-5.6 Sol |
|---|---|---|
| GPQA Diamond | 93.3% | 93.5% |
| SimpleQA Verified | 45.1% | 69.7% |
| Vectara Hallucination Rate | 7% | 12.4% |
| LMArena Expert | 1507 | 1516 |
| Humanity's Last Exam | 36.2% | — |
Multimodal GPT-5.6 Sol leads
GPT-5.4: 43.7 (#20), GPT-5.6 Sol: 48.6 (#9)
| Benchmark | GPT-5.4 | GPT-5.6 Sol |
|---|---|---|
| LMArena Vision | 1303 | 1281 |
| Blueprint-Bench 2 | 27.1% | 33.6% |
| Furniture Assembly | 37.5% | 56.7% |
| LMArena Document | 1471 | 1483 |
Multilingual Too close to call
GPT-5.4: 56.2 (#23), GPT-5.6 Sol: 55.3 (#32)
| Benchmark | GPT-5.4 | GPT-5.6 Sol |
|---|---|---|
| LMArena Non-English | 1465 | 1452 |
| LMArena Chinese | 1519 | 1527 |
| LMArena French | 1493 | 1477 |
| LMArena German | 1472 | 1476 |
| LMArena Japanese | 1485 | 1471 |
| LMArena Korean | 1448 | 1442 |
| LMArena Russian | 1480 | 1468 |
| LMArena Spanish | 1454 | 1441 |
Instruction Following Too close to call
GPT-5.4: 77.1 (#27), GPT-5.6 Sol: 77.7 (#16)
| Benchmark | GPT-5.4 | GPT-5.6 Sol |
|---|---|---|
| LMArena Instruction Following | 1469 | 1482 |
Long Context GPT-5.4 leads
GPT-5.4: 50.3 (#8), GPT-5.6 Sol: 45.4 (#42)
| Benchmark | GPT-5.4 | GPT-5.6 Sol |
|---|---|---|
| LMArena Longer Query | 1473 | 1480 |
| CL-bench | 27.9% | — |
| CL-bench Life | 21.7% | — |
Writing & Preference GPT-5.6 Sol leads
GPT-5.4: 71.9 (#17), GPT-5.6 Sol: 73.3 (#12)
| Benchmark | GPT-5.4 | GPT-5.6 Sol |
|---|---|---|
| LMArena Text | 1469 | 1457 |
| LMArena Creative Writing | 1439 | 1448 |
| EQ-Bench Creative Writing | 1840 | 1972 |
| EQ-Bench 4 | 1272 | 1250 |
| LMArena Multi-Turn | 1482 | 1460 |
Frequently asked questions
Is GPT-5.4 better than GPT-5.6 Sol?
GPT-5.6 Sol is the stronger model overall, scoring 65.0 to 59.4 on the Noometry Index.
Which is cheaper, GPT-5.4 or GPT-5.6 Sol?
GPT-5.4 is cheaper. It lists at $2.50 per million input tokens and $15 per million output tokens; GPT-5.6 Sol lists at $4 and $20.
Is GPT-5.4 or GPT-5.6 Sol better for coding?
GPT-5.6 Sol scores higher on coding benchmarks: 65.1 versus 52.6 in the Noometry coding category.
Which has the bigger context window?
Both accept 1.05M tokens.
How many benchmarks do GPT-5.4 and GPT-5.6 Sol share?
55 benchmarks have published results for both models. GPT-5.4 has 68 scored results on Noometry and GPT-5.6 Sol has 65.