Anthropic · Released Feb 4, 2026
Claude Opus 4.6
Opus 4.6 is Anthropic’s strongest model for coding and long-running professional tasks. It is built for agents that operate across entire workflows rather than single prompts, making it especially effective...
- #1DeepResearch Bench
- #1METR time horizon
- Top 12%SWE-bench Verified
- Top 13%Terminal-Bench
- Released
- Feb 4 2026
- Context window
- 1M 128K out
- Input / 1M tokens
- $5.00
- Output / 1M tokens
- $25.00
Where to run it
5 providers compared
Cheapest: Claude Platform on AWS, Google, Amazon Bedrock +2 more at $5.00 / $25.00 per 1M input / output tokens — 9% less than the priciest option.
| Provider | Input | Output | Cache read | Context | Max output | Uptime | vs. best |
|---|---|---|---|---|---|---|---|
| Claude Platform on AWS | $5.00 | $25.00 | $0.50 | 1M | 128K | 100%24h | Cheapest |
| Googleglobal | $5.00 | $25.00 | $0.50 | 1M | 128K | 100%24h | Cheapest |
| Amazon Bedrock | $5.00 | $25.00 | $0.50 | 1M | 128K | 100%24h | Cheapest |
| Anthropic | $5.00 | $25.00 | $0.50 | 1M | 128K | 100%24h | Cheapest |
| Azureglobal | $5.00 | $25.00 | $0.50 | 1M | 128K | — | Cheapest |
| Googleeurope | $5.50 | $27.50 | $0.55 | 1M | 128K | 100%24h | +10% |
Value
Price vs. performance
Claude Opus 4.6 Other Anthropic Other labs Value frontier · GPQA Diamond vs. input price per 1M tokens (log scale) · Drag to pan, pinch or Ctrl + scroll to zoom, double-click to zoom in.
Benchmarks
25 benchmark scores
| Benchmark | Score | Rank | Leader |
|---|---|---|---|
| FrontierCodeSoftware engineering · Mergeable fixes for hard open-source issues | 26.6% | #35of 41 | Claude Opus 5.554.6% |
| SWE-bench VerifiedSoftware engineering · Real GitHub issues fixed in Python repos | 78.7% | #3of 26 | Claude Opus 4.783.5% |
| Terminal-BenchSoftware engineering · Tasks completed with the programs in a real terminal (best agent) | 79.8% | #5of 40 | GPT-5.584.7% |
| GBAEvalSoftware engineering · Build a Game Boy Advance emulator from scratch | 44.1% | #11of 23 | Claude Opus 579.6% |
| GSOSoftware engineering · Optimize real code to make it significantly faster | 41.2% | #7of 20 | Claude Fable 5.188.2% |
| METR time horizonSoftware engineering · Length of software task solved more often than not | 12 h | #1of 13 | This model |
| APEX-AgentsAgents & computer use · Long professional tasks in banking, consulting and law | 46.3% | #33of 51 | Gemini 4 Argon82.4% |
| Remote Labor IndexAgents & computer use · Real paid freelance projects done end-to-end | 4.2% | #9of 13 | Claude Opus 5.521.3% |
| Vending-Bench 2Agents & computer use · Run a simulated vending business for a year (final balance) | $8,018 | #13of 61 | GPT-6 Astra$15,515 |
| DeepResearch BenchWriting & content · Find and synthesize information from the web | 55.3% | #1of 20 | This model |
| SimpleQA VerifiedWriting & content · Factual accuracy on 1,000 short questions | 47.0% | #31of 66 | Gemini 4 Argon76.0% |
| WebDev ArenaDesign & visual · People vote on which model builds the better website (Elo) | 1556 | #33of 88 | Claude Opus 5.51820 |
| Furniture AssemblyDesign & visual · Spot mistakes in photos of IKEA builds | 28.3% | #25of 34 | Claude Opus 5.583.3% |
| CybenchCybersecurity · Find and exploit vulnerabilities in sandboxed challenges | 93.0% | #1of 4 | This model |
| GPQA DiamondReasoning & knowledge · PhD-level biology, chemistry and physics questions | 90.5% | #35of 92 | Gemini 4 Argon96.4% |
| Humanity's Last ExamReasoning & knowledge · Expert questions across 100+ subjects | 34.4% | #9of 23 | GPT-6 Astra54.8% |
| ARC-AGI-2Reasoning & knowledge · Few-shot abstract reasoning on grid puzzles | 69.2% | #23of 63 | GPT-6 Astra95.0% |
| ARC-AGI-1Reasoning & knowledge · Pattern induction from input–output examples | 94.0% | #20of 63 | Gemini 4 Argon98.5% |
| SimpleBenchReasoning & knowledge · Common-sense trick questions about space, time and people | 67.6% | #13of 43 | Claude Fable 581.9% |
| Chess PuzzlesReasoning & knowledge · Find the single best move in 100 new puzzles | 21.0% | #56of 70 | GPT-6 Astra76.0% |
| LMCAReasoning & knowledge · Judge the quality of conceptual arguments like experts | 55.8% | #12of 96 | Claude Sonnet 5.568.3% |
| FrontierMathMaths · Expert-written problems, undergrad to early research (tiers 1–3) | 66.0% | #31of 72 | GPT-6.1 Sol93.7% |
| FrontierMath Tier 4Maths · Exceptionally hard research-level problems | 26.8% | #32of 62 | GPT-6.1 Sol100.0% |
| ProofBenchMaths · Graduate maths proofs verified in Lean 4 | 50.0% | #31of 70 | Claude Sonnet 5.5100.0% |
| OTIS Mock AIMEMaths · Competition-style problems harder than MATH Level 5 | 94.4% | #39of 90 | Gemini 4 Argon100.0% |
Specs
Details
- Model ID
anthropic/claude-opus-4.6- Developer
- Anthropic
- Release date
- Feb 4, 2026 ·
- Context window
- 1,000,000 tokens
- Max output
- 128,000 tokens
- API price
- $5.00 / $25.00 per 1M input / output tokens
- Providers
- 5
- Input types
- text, image, file
- Reasoning
- Yes
- Tool calling
- Yes
- Open weights
- No
- Free tier
- No
Release date, prices and providers from OpenRouter's catalog; buying directly from a provider can cost a little differently. Uptime is the share of OpenRouter requests each provider served over the last day. Benchmark scores are the best result per benchmark from Epoch AI, licensed CC BY 4.0. Rank counts every model tracked here with a score.