Claude Opus 4.8

2026-05-28 · Publisher : Anthropic · Domain : Artificial Intelligence

Anthropic launches Claude Opus 4.8, which extends Opus 4.7 with better results and more honest behavior on agentic tasks, at the same price, with 1M tokens of context by default.

What changes

  • Better judgment and honesty on agentic tasks; about four times fewer code flaws let through without comment than its predecessor, according to the vendor.
  • Dynamic workflows for large problems (research preview) and effort control in claude.ai and Cowork.
  • System messages accepted mid-conversation, without losing the prompt cache.
  • Fast mode (research preview) about 2.5 times faster.

### Specifications - Context window: 1M tokens by default. Maximum output: 128,000 tokens. Default effort: high. - Knowledge cutoff: not disclosed. Vision up to 2,576 pixels. - Availability: claude.ai, API, Bedrock, Google Cloud, Foundry. API identifier: `claude-opus-4-8`. ### Pricing | Item | Pricing (USD per million tokens) | | --- | --- | | Input | 5 | | Output | 25 | | Fast mode input / output | 10 / 50 | | Cache write 5 min / 1 h | 6.25 / 10 | | Cache read | 0.50 | | Batch (input / output) | 2.50 / 12.50 | ### Benchmarks The announcement's comparison table is an image; only a few figures appear in the text (scores published by the vendor or quoted from partners): Online-Mind2Web 84% (partner testimonial), first model to exceed 10% on the Legal Agent Benchmark. Footnote: Opus 4.7 updated to 82.3% on OSWorld-Verified. ### Key takeaways Opus 4.8 serves as the fallback for Fable 5's safety classifiers and remains the Opus reference at $5 / $25 until Opus 5.

Links

Tags : LLM, Anthropic, Claude, Opus, Agents, Contexte 1M