Anthropic ships Claude Opus 4.8 with honesty over benchmarks

Anthropic released Claude Opus 4.8 just 41 days after Opus 4.7, calling it a "modest but tangible improvement." The model is four times better at flagging code flaws, fast mode is three times cheaper at $10 per million input tokens, and the prompt cache minimum dropped from 4,096 to 1,024 tokens.

• SWE-bench Pro rose from 67.1% to 69.2%
• Fast mode costs $10 per million input tokens vs $30
• Model is 4x better at flagging code flaws
• Dynamic Workflows runs hundreds of parallel subagents