Anthropic's most capable public model. Best-in-class coding, 1M context, hybrid reasoning — and the same $5/$25 pricing as its predecessor.
Released April 16, 2026 — the biggest Opus upgrade since 4.6.
SWE-bench Pro score jumped from 53.4% → 64.3%, overtaking GPT-5.4 (57.7%). On a 93-task internal benchmark, Opus 4.7 resolved 13% more tasks than its predecessor — including four that no prior Claude model could solve.
Expanded from 200K to 1 million tokens — fit entire codebases, full legal documents, or hours of transcripts into a single request without chunking.
A new reasoning effort setting between high and max. Lets you dial up reasoning depth on hard problems without paying the full latency cost of max effort.
Vision accuracy jumped from 54.5% → 98.5% — near-perfect. Combined with 3× higher resolution (2,576px, ~3.75MP). Diagrams, scanned contracts, financial tables, and screenshots are now read reliably.
Better long-horizon autonomy, improved systems engineering, and new task budgets and Claude Code review tools make Opus 4.7 the clearest choice for autonomous coding agents.
Opus 4.7 uses a new tokenizer that encodes the same text into up to 1.35× more tokens. Per-token prices are the same as Opus 4.6 — but actual costs per request can be meaningfully higher on long prompts.
Head-to-head on the metrics that matter most.
| Metric | Opus 4.7 | Opus 4.6 | GPT-5.4 |
|---|---|---|---|
| SWE-bench Pro (coding) | 64.3% | 53.4% | 57.7% |
| SWE-bench Verified | 87.6% | 80.8% | ~80% |
| CursorBench (agents) | 70% | 58% | 58% |
| GPQA Diamond (reasoning) | 94.2% | ~90% | 94.4% |
| Vision accuracy | 98.5% | 54.5% | Lower |
| Context window | 1M tokens | 200K tokens | 128K tokens |
| Vision resolution | 2,576px (~3.75MP) | ~800px | 2,048px |
| Input pricing | $5/1M | $5/1M | $0.75/1M |
| Output pricing | $25/1M | $25/1M | $4.50/1M |
SWE-bench Pro scores as of April 16, 2026. Pricing in USD per million tokens at standard API rates.
Best-in-class coding: 64.3% on SWE-bench Pro, #1 across all public models
Massive 1M token context window — handles entire codebases in one shot
Hybrid reasoning with new xhigh effort for deeper problem solving
Vision accuracy jumped from 54.5% → 98.5% — near-perfect on charts, docs, screenshots
Vision up to 2,576px long edge (~3.75 MP) — 3× higher resolution than prior Claude
Best for long-running agents: systems engineering, complex multi-step tasks
Strong at professional document work: slides, financial analysis, data visualization
90% cost reduction with prompt caching; 50% with batch processing
Available on every major cloud: AWS Bedrock, Google Vertex, Microsoft Foundry
New tokenizer maps text to up to 1.35× more tokens — real cost can exceed Opus 4.6
Slower than Claude Sonnet 4.6 — not ideal for high-volume or latency-sensitive apps
No built-in image generation (text model only)
Max effort settings add latency — not suitable for real-time chat at scale
Overkill for simple tasks: writing emails, basic Q&A, short summaries
Same per-token rates as Opus 4.6 — but the new tokenizer changes your effective cost.
Tokenizer note: Opus 4.7 uses a new tokenizer that can encode the same prompt into up to 1.35× more tokens than Opus 4.6. Your actual spend per request may be higher even though the per-token price is unchanged. Test your specific prompts before migrating production workloads.
See how it fits across the full model landscape.
Claude Opus 4.7 is Anthropic's most capable publicly available model, released April 16 2026. It's a hybrid reasoning model with a 1 million token context window, improved coding and agent capabilities, and a new xhigh effort level. It's the second most powerful Anthropic model overall — after Claude Mythos, which is restricted to select research partners.
Claude Opus 4.7 is priced at $5 per million input tokens and $25 per million output tokens — the same nominal price as Opus 4.6. However, Opus 4.7 uses a new tokenizer that maps text to up to 1.35× more tokens, so real-world costs per request can be higher. Prompt caching cuts costs by up to 90%, and batch processing saves 50%.
Claude Opus 4.7 leads on SWE-bench Pro with a score of 64.3%, ahead of GPT-5.4 at 57.7%. On a 93-task coding benchmark it resolved 13% more tasks than Opus 4.6, including four tasks that neither Opus 4.6 nor Sonnet 4.6 could solve.
Upgrade if you run complex agentic coding workflows, work with large images, or need the deepest long-horizon reasoning available. Stay on Opus 4.6 if your tasks are straightforward and you're sensitive to token cost — the new tokenizer means identical prompts can cost up to 35% more with Opus 4.7.
xhigh is a new reasoning effort setting that sits between the existing high and max levels. It gives developers finer control over the tradeoff between reasoning depth and latency on hard problems — useful when you need deeper thinking than high but don't want the latency of max.
Claude Opus 4.7 is available on claude.ai (Pro and Max subscription plans), the Anthropic API, AWS Bedrock, Google Vertex AI, and Microsoft Foundry (Azure). The model ID is claude-opus-4-7-20260416.
Claude Mythos is Anthropic's most powerful model overall but is not publicly available — it's currently limited to 11 organizations for cybersecurity research. Opus 4.7 is the most capable model you can actually use today. Notably, Anthropic deliberately reduced cyber capabilities in Opus 4.7 compared to what Mythos can do.
