Model comparison
Gemini 3.1 Pro Preview vs GPT-5.4
GPT-5.4 is the stronger model overall, scoring 59.4 to 56.7 on the Noometry Index.
Last verified . 67 shared benchmarks.
GoogleSummary
- They share 67 benchmarks with published results for both. Gemini 3.1 Pro Preview scores higher in 3 categories and GPT-5.4 in 7 categories; 8 gaps are clear of the uncertainty.
- The widest gap is in math, where GPT-5.4 leads 73.5 to 62.1.
- The biggest single-benchmark swing is GBAEval: 0.8% for Gemini 3.1 Pro Preview and 45.1% for GPT-5.4.
- Gemini 3.1 Pro Preview is cheaper at $2 / $12 per million input/output tokens, against $2.50 / $15 for GPT-5.4.
- GPT-5.4 accepts more context: 1.05M tokens versus 1.05M.
Side by side
| Gemini 3.1 Pro Preview | GPT-5.4 | |
|---|---|---|
| Provider | OpenAI | |
| Noometry Index | 56.7 | 59.4 |
| Released | 2026-02-19 | 2026-03-05 |
| Weights | Proprietary | Proprietary |
| Context window | 1.05M | 1.05M |
| Max output | 66K | 128K |
| Input $ / M tokens | $2 | $2.50 |
| Output $ / M tokens | $12 | $15 |
| Results tracked | 71 | 68 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding GPT-5.4 leads
Gemini 3.1 Pro Preview: 42.5 (#99), GPT-5.4: 52.6 (#33)
| Benchmark | Gemini 3.1 Pro Preview | GPT-5.4 |
|---|---|---|
| SWE-bench Verified | 75.6% | 76.9% |
| DeepSWE | 11.7% | 51.8% |
| LMArena WebDev | 1447 | 1465 |
| SciCode | 58.9% | 56.6% |
| GSO | 22.6% | 31.4% |
| WeirdML | 72.1% | 77.7% |
| LMArena Coding | 1484 | 1497 |
| MirrorCode | 8.9% | 15.6% |
| ALE-Bench | 1,161 | 1,607 |
| AlgoTune | 2.02 | 1.85 |
Agentic & Tool Use GPT-5.4 leads
Gemini 3.1 Pro Preview: 37.7 (#34), GPT-5.4: 46.5 (#13)
| Benchmark | Gemini 3.1 Pro Preview | GPT-5.4 |
|---|---|---|
| Terminal-Bench | 80.2% | 81.8% |
| APEX-Agents | 35.3% | 52.4% |
| τ²-bench Banking | 26% | 39.4% |
| DeepResearch Bench | 47.8% | 35.1% |
| PostTrainBench | 22% | 19% |
| GBAEval | 0.8% | 45.1% |
| LMArena Search | 1211 | 1197 |
| METR Time Horizons | 77% | 74.3% |
| Vending-Bench 2 | 3,774 | 6,144 |
| BALROG | 57% | — |
| ExploitBench | 26.1% | — |
| GDP.pdf | 17% | — |
Reasoning Gemini 3.1 Pro Preview leads
Gemini 3.1 Pro Preview: 71.7 (#12), GPT-5.4: 61.8 (#19)
| Benchmark | Gemini 3.1 Pro Preview | GPT-5.4 |
|---|---|---|
| ARC-AGI-2 | 77.1% | 74% |
| NYT Connections (extended) | 97.4% | 91.3% |
| ARC-AGI-1 | 98% | 93.7% |
| CritPt | 17.7% | 23.4% |
| Chess Puzzles | 55% | 44% |
| EnigmaEval | 36.8% | 16% |
| Thematic Generalization | 79.4% | 80% |
| EBR-Bench | 14.3% | 25.4% |
| LMArena Hard Prompts | 1485 | 1485 |
| Mystery Game Puzzles | 34% | 37% |
| DTBench | 97.1% | 94.4% |
| LMCA | 53.8% | 52% |
| Epoch Capabilities Index | 154.77 | 156.81 |
| ForecastBench | 59 | 59.5 |
| SimpleBench | 79.6% | — |
| Kagi LLM Benchmark | — | 63.8% |
Math GPT-5.4 leads
Gemini 3.1 Pro Preview: 62.1 (#34), GPT-5.4: 73.5 (#19)
| Benchmark | Gemini 3.1 Pro Preview | GPT-5.4 |
|---|---|---|
| FrontierMath (Tiers 1-3) | 59.6% | 78.6% |
| FrontierMath Tier 4 | 26.8% | 49% |
| MathArena Final-Answer Competitions | 86.5% | 83.1% |
| OTIS Mock AIME 2024-2025 | 95.6% | 97.8% |
| ProofBench | 26% | 56% |
| LMArena Math | 1485 | 1488 |
| FrontierMath (Feb 2025 set) | 36.9% | 47.6% |
| FrontierMath Tier 4 (v1) | 16.7% | 27.1% |
Knowledge Gemini 3.1 Pro Preview leads
Gemini 3.1 Pro Preview: 71.8 (#3), GPT-5.4: 65.3 (#14)
| Benchmark | Gemini 3.1 Pro Preview | GPT-5.4 |
|---|---|---|
| GPQA Diamond | 94.4% | 93.3% |
| Humanity's Last Exam | 46.4% | 36.2% |
| SimpleQA Verified | 73.5% | 45.1% |
| Vectara Hallucination Rate | 10.4% | 7% |
| LMArena Expert | 1485 | 1507 |
Multimodal GPT-5.4 leads
Gemini 3.1 Pro Preview: 37.9 (#69), GPT-5.4: 43.7 (#20)
| Benchmark | Gemini 3.1 Pro Preview | GPT-5.4 |
|---|---|---|
| LMArena Vision | 1296 | 1303 |
| Blueprint-Bench 2 | 26.5% | 27.1% |
| Furniture Assembly | 26.7% | 37.5% |
| LMArena Document | 1444 | 1471 |
Multilingual Too close to call
Gemini 3.1 Pro Preview: 57.0 (#12), GPT-5.4: 56.2 (#23)
| Benchmark | Gemini 3.1 Pro Preview | GPT-5.4 |
|---|---|---|
| LMArena Non-English | 1477 | 1465 |
| LMArena Chinese | 1529 | 1519 |
| LMArena French | 1487 | 1493 |
| LMArena German | 1491 | 1472 |
| LMArena Japanese | 1493 | 1485 |
| LMArena Korean | 1455 | 1448 |
| LMArena Russian | 1498 | 1480 |
| LMArena Spanish | 1479 | 1454 |
Instruction Following Too close to call
Gemini 3.1 Pro Preview: 77.0 (#32), GPT-5.4: 77.1 (#27)
| Benchmark | Gemini 3.1 Pro Preview | GPT-5.4 |
|---|---|---|
| LMArena Instruction Following | 1466 | 1469 |
Long Context GPT-5.4 leads
Gemini 3.1 Pro Preview: 47.4 (#18), GPT-5.4: 50.3 (#8)
| Benchmark | Gemini 3.1 Pro Preview | GPT-5.4 |
|---|---|---|
| CL-bench | 20.8% | 27.9% |
| CL-bench Life | 16.9% | 21.7% |
| LMArena Longer Query | 1483 | 1473 |
Writing & Preference GPT-5.4 leads
Gemini 3.1 Pro Preview: 66.1 (#37), GPT-5.4: 71.9 (#17)
| Benchmark | Gemini 3.1 Pro Preview | GPT-5.4 |
|---|---|---|
| LMArena Text | 1481 | 1469 |
| LMArena Creative Writing | 1482 | 1439 |
| EQ-Bench Creative Writing | 1491 | 1840 |
| EQ-Bench 4 | 1142 | 1272 |
| LMArena Multi-Turn | 1488 | 1482 |
Frequently asked questions
Is Gemini 3.1 Pro Preview better than GPT-5.4?
GPT-5.4 is the stronger model overall, scoring 59.4 to 56.7 on the Noometry Index.
Which is cheaper, Gemini 3.1 Pro Preview or GPT-5.4?
Gemini 3.1 Pro Preview is cheaper. It lists at $2 per million input tokens and $12 per million output tokens; GPT-5.4 lists at $2.50 and $15.
Is Gemini 3.1 Pro Preview or GPT-5.4 better for coding?
GPT-5.4 scores higher on coding benchmarks: 52.6 versus 42.5 in the Noometry coding category.
Which has the bigger context window?
GPT-5.4 does, with 1.05M tokens against 1.05M.
How many benchmarks do Gemini 3.1 Pro Preview and GPT-5.4 share?
67 benchmarks have published results for both models. Gemini 3.1 Pro Preview has 71 scored results on Noometry and GPT-5.4 has 68.