Compare / head-to-head
Claude Sonnet 4.6vs
GPT-5.3-Codex
Claude Sonnet 4.6 leads 3 of 5 shared benchmarks. GPT-5.3-Codex is 1.1x cheaper per token. Claude Sonnet 4.6 has the larger context window (1M tokens).
Benchmarks from cited public sources; pricing from official pages; status from official provider feeds.
Shared benchmarks
3 – 2
Claude Sonnet 4.6 leads
Cheaper per token
GPT-5.3-Codex
1.1x cheaper, input + output
Larger context
Claude Sonnet 4.6
1M tokens
Providers
2 providers
Anthropic · OpenAI
Claude Sonnet 4.6
Anthropic · released 2026-02-17
textvision
GPT-5.3-Codex
OpenAI · released 2026-02-05
textvision
Quality
Benchmark matrix
5 shared · 30 only Claude Sonnet 4.6 · 4 only GPT-5.3-Codex| Benchmark | Claude Sonnet 4.6 | GPT-5.3-Codex | Δ | Edge |
|---|---|---|---|---|
| Reported by both · 5 | ||||
| OSWorld-Verified | 72.5% ↗ | 64.7% ↗ | +7.8 pt | Claude Sonnet 4.6 |
| SWE-bench Verified (Epoch AI run) | 75.2% ↗epoch run | 74.8% ↗epoch run | +0.4 pt | Claude Sonnet 4.6 |
| Terminal-Bench 2.0 | 59.1% ↗ | 77.3% ↗ | -18.2 pt | GPT-5.3-Codex |
| Vending-Bench 2 (Andon Labs) | 7204.14 ↗ | 5940.12 ↗ | +1264 | Claude Sonnet 4.6 |
| WeirdML (Håvard Tveit Ihle) | 66.1% ↗ | 79.3% ↗ | -13.2 pt | GPT-5.3-Codex |
| Only Claude Sonnet 4.6 reports · 30 | ||||
| SWE-bench Verified | 79.6% ↗ | not reported | — | — |
| GPQA Diamond | 89.9% ↗ | not reported | — | — |
| Humanity's Last Exam (no tools) | 33.2% ↗ | not reported | — | — |
| LMArena Elo | 1458.3 ↗ | not reported | — | — |
| APEX-Agents (Mercor) | 43% ↗ | not reported | — | — |
| ARC-AGI-2 (Verified) | 58.3% ↗ | not reported | — | — |
| Chess Puzzles (Epoch AI run) | 13% ↗epoch run | not reported | — | — |
| DeepSWE v1.1 (Datacurve) | 29.9% ↗ | not reported | — | — |
| GDPval-AA | 1633 ↗ | not reported | — | — |
| Humanity's Last Exam (with tools) | 49% ↗ | not reported | — | — |
| LiveBench Agentic Coding (LiveBench) | 42.6% ↗ | not reported | — | — |
| LiveBench Coding (LiveBench) | 79.3% ↗ | not reported | — | — |
| LiveBench Data Analysis (LiveBench) | 77.9% ↗ | not reported | — | — |
| LiveBench Instruction Following (LiveBench) | 63.2% ↗ | not reported | — | — |
| LiveBench Language (LiveBench) | 76.1% ↗ | not reported | — | — |
| LiveBench Mathematics (LiveBench) | 87% ↗ | not reported | — | — |
| LiveBench Reasoning (LiveBench) | 84.8% ↗ | not reported | — | — |
| LMArena Vision (LMArena) | 1275.2 ↗ | not reported | — | — |
| LMArena WebDev (LMArena) | 1521.4 ↗ | not reported | — | — |
| MCP Atlas | 61.3% ↗ | not reported | — | — |
| MMMLU | 89.3% ↗ | not reported | — | — |
| MMMU-Pro (no tools) | 74.5% ↗ | not reported | — | — |
| MMMU-Pro (with tools) | 75.6% ↗ | not reported | — | — |
| Mystery Game Puzzles (Epoch AI run) | 16% ↗epoch run | not reported | — | — |
| OSWorld-Verified (XLANG) | 72.1% ↗ | not reported | — | — |
| OTIS Mock AIME 2024-2025 (Epoch AI run) | 85.8% ↗epoch run | not reported | — | — |
| SAGE (Vals AI) | 46.6% ↗ | not reported | — | — |
| SimpleQA Verified | 35.5% ↗epoch run | not reported | — | — |
| Tau2-bench Retail | 91.7% ↗ | not reported | — | — |
| Tau2-bench Telecom | 97.9% ↗ | not reported | — | — |
| Only GPT-5.3-Codex reports · 4 | ||||
| Cybersecurity Capture The Flag Challenges | not reported | 77.6% ↗ | — | — |
| GDPval (wins or ties) | not reported | 70.9% ↗ | — | — |
| SWE-Bench Pro (Public) | not reported | 56.8% ↗ | — | — |
| SWE-Lancer IC Diamond | not reported | 81.4% ↗ | — | — |
Scores tagged "epoch" or "matharena" are independent runs, used only where the lab has not published its own; ⚠ marks rows MathArena flags as released after the competition. Higher is better on every row. Δ is Claude Sonnet 4.6 minus GPT-5.3-Codex in the benchmark's own unit. "Not reported" means the lab has not published that figure; it is not a zero. ↗ opens the source.
Specs & pricing
Side by side
Official pricing pages and model cards| Spec | Claude Sonnet 4.6 | GPT-5.3-Codex | Edge |
|---|---|---|---|
| Context window | 1M tokens | 400K tokens | Claude Sonnet 4.6 |
| Max output | 128K tokens | 128K tokens | Tie |
| Input price / 1M | $3 ↗ | $1.75 ↗ | GPT-5.3-Codex |
| Output price / 1M | $15 ↗ | $14 ↗ | GPT-5.3-Codex |
| Cached input / 1M | $0.3 ↗ | $0.175 ↗ | GPT-5.3-Codex |
| Input + output / 1M Lower is cheaper. List prices; batch, tool and regional fees excluded. | $18 | $15.75 | GPT-5.3-Codex |
| Modalities | text · vision | text · vision | Tie |
| Released | 2026-02-17 | 2026-02-05 | — |
| Cited benchmark scores | 41 | 9 | — |
Reliability
Provider status
Live from /statusMore matchups
Claude Sonnet 4.6 vs …
Models sharing the most benchmarksMore matchups
GPT-5.3-Codex vs …
Models sharing the most benchmarksBuilt by Respan
Which one wins on your data?
Public benchmarks are a starting point. Run Claude Sonnet 4.6 and GPT-5.3-Codex on your own prompts with Respan evals, or route to either through one gateway key with automatic failover.