GPT-5.4
OpenAIProprietaryReasoningVerifiedReleased Mar 5, 2026Input: text, image, file
Overall score
66.7
Rank 30 of 222 · 62.3–71.2
Context window
1.05M
1,050,000 tokens
Price per 1M tokens
$2.50 in · $15.00 out
✓ Verified on OpenAI's pricing page
Blended price
$5.63
3:1 input:output tokens
Output speed
—
Not measured yet
Category scores
Scores per capability area with the model's rank among ranked models in that category.
- 1Instruction following#2390.7
- 2Math82.0
- 3Agentic & tool use#4580.3
- 4Knowledge#2177.3
- 5Multimodal#3173.1
- 6Reasoning#2064.2
- 7Coding#4947.1
Benchmark results (40)
As reported in system cards, papers and public evaluations. Position compares this model with every other model reporting the same benchmark.
Agentic & tool use
| Gert Labs | 64.89 | #4 of 52 |
| BrowseComp | 82.7 | #22 of 50 |
| MCP Atlas | 70.6 | #24 of 42 |
| MazeBench | 0 | #20 of 42 |
| Terminal-Bench 2.0 | 75.1 | #3 of 39 |
| Claw-Eval | 60.3 | #22 of 36 |
| OSWorld-Verified | 75.0 | #16 of 35 |
| JobBench | 38.9 | #17 of 34 |
| CyberGym | 79.0 | #14 of 25 |
| Toolathlon | 54.6 | #6 of 22 |
| DeepSearchQA | 73.6 | #16 of 20 |
| ResearchClawBench | 15.3 | #13 of 19 |
| ApprenticeBench | 11 | #14 of 18 |
| ExploitGym | 6.0 | #12 of 14 |
Multimodal
| Design Arena Website | 1221 | #54 of 91 |
| MMMU-Pro | 81.2 | #7 of 48 |
| CharXiv | 82.8 | #17 of 42 |
| ParseBench | 62.23 | #25 of 37 |
| SimpleVQA | 61.1 | #12 of 17 |
| ERQA | 65.4 | #10 of 16 |
| ScreenSpot Pro | 85.4 | #3 of 15 |
| OfficeQA Pro | 53.2 | #11 of 13 |
| MMMU-Pro w/ Python | 82.1 | #4 of 9 |
| ZeroBench | 41.0 | #2 of 8 |
| MedXpertQA (MM) | 77.1 | #4 of 8 |
Knowledge
| GPQA | 92.8 | #9 of 117 |
| HLE | 52.1 | #16 of 69 |
| HLE w/o tools | 39.8 | #24 of 43 |
| HealthBench Professional | 48.1 | #16 of 16 |
| HealthBench Hard | 40.1 | #2 of 12 |
| MedXpertQA (Text) | 59.6 | #2 of 5 |
Coding
| SWE-bench Pro | 57.7 | #34 of 69 |
| React Native Evals | 85.3 | #2 of 15 |
| PostTrainBench v1.1 | 19.0 | #15 of 15 |
| LiveCodeBench Pro | 87.5 | #1 of 6 |
Reasoning
| ARC-AGI-2 | 74.0 | #19 of 41 |
| ARC-AGI-1 | 93.67 | #17 of 36 |
| ARC-AGI-3 | 0.2 | #17 of 22 |
Math
| FrontierMath v2 (Tiers 1-3) | 78.596 | #18 of 64 |
| FrontierMath v2 (Tier 4) | 49.000 | #18 of 51 |
Pricing by source
USD per 1M tokens.
| Source | Input | Output | Cached in |
|---|---|---|---|
| ✓ Official (OpenAI) checked Oct 17, 2026 · gpt-5.4 | $2.50 | $15.00 | $0.25 |
| BenchLM list price | $2.50 | $15.00 | $0.25 |
| OpenRouter openai/gpt-5.4 | $2.50 | $15.00 | $0.25 |
Standard tier, prompts under 272K tokens.
OpenAI's current API pricing page lists GPT-5.4 at $2.50 input / $0.25 cached input / $15.00 output per million tokens for short-context requests, with higher pricing for long-context requests.
Arena ratings
Human-preference ratings from LMArena. Best-rated variant shown.
| Text(high) | 1469 | #33 · 64.8K votes |
| WebDev (coding)(High, Codex Harness) | 1464 | #72 · 1.4K votes |
| Vision(High) | 1303 | #18 · 30.7K votes |
| Documents | 1471 | #14 · 33.3K votes |
Speed
Measured by us through OpenRouter (median of trials).
Not measured yet.
Related models
Same family
- GPT-5.4 Pro68.7
- GPT-5.4 mini53.4
- GPT-5.4 nano46.9
Similar score
- GPT-5.2 Pro66.7 · $57.75
- GPT-6 Luna66.6 · $0.20
- Gemini 3.7 Flash66.9 · $1.50
- Grok 4.667.1 · $3.00
- MiniMax M3.1 Flash67.1 · $0.26