Claude Opus 4.5
2025-11-24 · Publisher : Anthropic · Domain : Artificial Intelligence
Anthropic launches Claude Opus 4.5, its most capable model in coding, agents and computer use, with a price cut by two thirds (5 / 25 USD) and a new effort parameter.
What changes
- New effort parameter to trade off capability against token consumption.
- At medium effort, matches Sonnet 4.5's best SWE-bench Verified score with 76% fewer output tokens.
- At high effort, exceeds Sonnet 4.5 by 4.3 points with 48% fewer tokens.
- Aider Polyglot: +10.6% compared with Sonnet 4.5. Vending-Bench: +29%. Terminal Bench (Warp): +15%.
- Prompt injection resistance higher than other frontier models according to the vendor; scores higher than any human candidate on Anthropic's internal technical exam (2 h, parallel computing).
- Context and memory management for agents; long conversations automatically summarized in the apps.
### Specifications - API identifier: claude-opus-4-5-20251101 (alias claude-opus-4-5). - 200K-token context, 64K-token maximum output. Reliable knowledge cutoff: May 2025; training data through August 2025. - Availability: Claude apps, API and the three major clouds (including Bedrock, Vertex AI, Microsoft Foundry). ### Pricing | Input (USD / M tokens) | Output (USD / M tokens) | Cache write 5 min / 1 h | Cache read | Batch | |---|---|---|---|---| | 5 | 25 | 6.25 / 10 | 0.50 | 2.50 / 12.50 | Versus 15 / 75 USD for Opus 4 and 4.1. ### Benchmarks | Benchmark | Result | |---|---| | SWE-bench Verified | best of the frontier models compared, exact score not recorded | | SWE-bench Multilingual | first on 7 of 8 languages | | Aider Polyglot | +10.6% vs Sonnet 4.5 | | Vending-Bench | +29% vs Sonnet 4.5 | Scores published by the vendor. ### Key takeaways Opus 4.5 makes Opus-level capability available at a third of the previous price, with an effort setting to control the cost of agentic tasks.
Links
Tags : LLM, Anthropic, Claude, Opus, Agents, Codage, Effort