Anthropic has officially launched Claude Opus 4.7, its most sophisticated and capable AI model to date. Released yesterday, the new model is designed specifically for high-stakes enterprise workflows, advanced software engineering, and complex, multi-step agentic tasks.
While it is the current public flagship, Anthropic notably confirmed that Opus 4.7 is “less capable” than Claude Mythos, a restricted model the company deemed too advanced for general public release due to cybersecurity and safety considerations.
Key Advancements in Opus 4.7
Opus 4.7 introduces a “hybrid reasoning” architecture that allows it to self-verify its work before delivering a final response.
- Autonomous Software Engineering: In internal tests, Opus 4.7 built a complete Rust-based text-to-speech engine from scratch, including neural models and SIMD kernels, and verified its own output against a reference.
- High-Resolution Multimodal Vision: The model now supports images up to 2,576 pixels on the long edge (~3.75 megapixels), a 3x improvement over prior versions. This enables pixel-perfect analysis of dense screenshots, complex UI diagrams, and technical blueprints.
- Adaptive Thinking & “Extra High” Effort: A new “xhigh” (extra high) effort setting allows users to trade latency for deeper reasoning on “impossible” problems.
- 1-Million Token Context: The model maintains high performance and instruction-following across its full 1M token window, making it ideal for analyzing massive codebases or multi-day legal projects.
Performance Benchmarks
Opus 4.7 has reclaimed the top spot on several key leaderboards, particularly in the “humanity” and coding categories:
| Benchmark | Claude Opus 4.7 | Gemini 3.1 Pro | GPT-5-4 Pro |
| Humanity’s Last Exam (No Tools) | 46.9% | 44.4% | 42.7% |
| SWE-bench Pro (Coding) | 64.3% | Pending | 61.2% |
| Finance Agent v1.1 | 64.4% | 62.1% | 63.8% |
Pricing and Tokenizer “Sticker Shock”
While Anthropic has kept the official sticker price identical to Opus 4.6, developers are noticing a change in actual costs due to a new tokenizer.
- List Price: $5 per 1M input tokens / $25 per 1M output tokens.
- The Token Inflation: The new tokenizer produces 1.0x to 1.35x more tokens for the same amount of text. This means that while the rate is the same, the bill for a specific prompt could be up to 35% higher.
- Savings: Anthropic continues to offer up to 90% discounts for prompt caching and 50% for batch processing.
Availability
Claude Opus 4.7 is available starting today via:
- Claude.ai: For Pro, Max, and Team subscribers.
- API Platforms: Native Anthropic Console, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry.
- Claude Code: The CLI tool has been updated to default to the “xhigh” effort level for all premium users.
Safety Note: Despite its power, Anthropic reports that Opus 4.7 is “more reliably honest” than its predecessors, with a significant reduction in hallucination rates and a stronger resistance to “jailbreak” or prompt injection attacks.
