GPT-5.4vsClaude Opus 4.6
Across 12 shared benchmarks, GPT-5.4 leads overall: GPT-5.4 wins 8, Claude Opus 4.6 wins 3, with 1 ties and an average score difference of +0.25.
GPT-5.4
OpenAI · 2026-03-05 · Multimodal model
Across 12 shared benchmarks, GPT-5.4 leads overall: GPT-5.4 wins 8, Claude Opus 4.6 wins 3, with 1 ties and an average score difference of +0.25.
OpenAI · 2026-03-05 · Multimodal model
Anthropic · 2026-02-05 · Reasoning model
Grouped by capability, sorted by largest gap within each. 12 shared benchmarks.
| Benchmark | GPT-5.4 | Claude Opus 4.6 | Diff |
|---|---|---|---|
| ARC-AGI-2 | 77.107 / 58Normal (No Tools) | 66.3014 / 58Extended (no tools) | +10.80 |
| ARC-AGI | 93.707 / 65Normal (No Tools) | 9211 / 65Extended (no tools) | +1.70 |
| GPQA Diamond | 92.809 / 175极高强度思考（无工具） | 91.3112 / 175Extended (no tools) | +1.49 |
| HLE | 52.1011 / 149极高强度思考（工具） | 538 / 149Extended (with tools, internet) | -0.90 |
| ARC-AGI-3 | 04 / 6Thinking High (No Tools) | 01 / 6最高（无工具） | — |
| Benchmark | GPT-5.4 | Claude Opus 4.6 | Diff |
|---|---|---|---|
| Terminal Bench 2.0 | 75.104 / 43极高强度思考（工具） | 65.409 / 43Extended (with tools) | +9.70 |
| OSWorld-Verified | 754 / 14极高强度思考（工具） | 72.706 / 14Extended (with tools) | +2.30 |
| Benchmark | GPT-5.4 | Claude Opus 4.6 | Diff |
|---|---|---|---|
| FrontierMath | 47.605 / 60极高强度思考（无工具） | 40.707 / 60最高（无工具） | +6.90 |
| FrontierMath - Tier 4 | 27.1011 / 80极高强度思考（无工具） | 22.9012 / 80最高（无工具） | +4.20 |
| Benchmark | GPT-5.4 | Claude Opus 4.6 | Diff |
|---|---|---|---|
| τ²-Bench - Telecom | 64.3030 / 35Normal (With Tools) | 99.252 / 35Extended (with tools) | -34.95 |
| Benchmark | GPT-5.4 | Claude Opus 4.6 | Diff |
|---|---|---|---|
| BrowseComp | 82.709 / 43极高强度思考（工具） | 846 / 43Thinking (With Tools + Internet) | -1.30 |
| Benchmark | GPT-5.4 | Claude Opus 4.6 | Diff |
|---|---|---|---|
| Pinch Bench | 90.501 / 37Thinking (With Tools) | 87.407 / 37Thinking (With Tools) | +3.10 |
| Field | GPT-5.4 | Claude Opus 4.6 |
|---|---|---|
| Publisher | OpenAI | Anthropic |
| Release date | 2026-03-05 | 2026-02-05 |
| Model type | Multimodal model | Reasoning model |
| Architecture | Dense | Dense |
| Parameters | 0.0 | 0.0 |
| Context length | 1M | 1000K |
| Max output | 128000 | 65536 |
Prices use DataLearner records when available; missing fields are not inferred.
| Item | GPT-5.4 | Claude Opus 4.6 |
|---|---|---|
| Text input | $2.5 / 1M tokens | $0.5 / 1M tokens |
| Text output | $15 / 1M tokens | $25 / 1M tokens |
| Cache read | Not public | $0.5 / 1M tokens |
| Cache write | $0.25 / 1M tokens | $10 / 1M tokens |
On average across the 12 shared benchmarks, GPT-5.4 scores 0.25 higher.
Largest single-benchmark gap: τ²-Bench - Telecom — GPT-5.4 64.30 vs Claude Opus 4.6 99.25 (-34.95).
Page generated from structured model, pricing and benchmark records. No real-time LLM is used to write the prose.
