Claude Opus 4.8 ships — same price, more honest, Fast Mode cost cut 3×
TL;DR
Claude Opus 4.8 ships with Anthropic's H-round announcement. Same API price, beats Opus 4.7 across coding, agents, reasoning. 4× more likely to flag code bugs honestly. Fast Mode unit cost cut 3×.
Today (May 28) Anthropic shipped Claude Opus 4.8 alongside its H-round announcement. Across coding, agent tasks, reasoning, and knowledge work benchmarks, Opus 4.8 beats Opus 4.7. The most prominent improvement is honesty — Opus 4.8 is 4× more likely than its predecessor to flag a problem in code rather than let the bug slip through. Early testers report it as more reliable, better-judged, and able to hold direction across long tasks.
On legal benchmarks, Opus 4.8 is the first model crossing 10% on the all-pass standard. On computer-use and browser agent eval Online-Mind2Web, scores 84% — clearly above Opus 4.7 and GPT-5.5.
Pricing unchanged, Fast Mode 3× cheaper. API: input $5 / M tokens, output $25 / M. Fast Mode (2.5× speed): input $10, output $50 — but Fast Mode unit cost is cut 3× vs prior generation.
Three new features same day. Dynamic Workflows hits Claude Code research preview — parallel hundreds of subagents handling repo-scale migrations. claude.ai and Cowork gain effort level controls — pick Claude's thinking depth, higher effort = better outcomes for faster rate-limit burn. The API now accepts system entries directly inside the messages array — let developers update instructions mid-agent-run without breaking prompt cache.
Next: Anthropic confirms it's building a new model class above Opus — the Mythos tier. Still in limited testing in cybersecurity. Expected to broaden to all users within weeks.
via Anthropic
On legal benchmarks, Opus 4.8 is the first model crossing 10% on the all-pass standard. On computer-use and browser agent eval Online-Mind2Web, scores 84% — clearly above Opus 4.7 and GPT-5.5.
Pricing unchanged, Fast Mode 3× cheaper. API: input $5 / M tokens, output $25 / M. Fast Mode (2.5× speed): input $10, output $50 — but Fast Mode unit cost is cut 3× vs prior generation.
Three new features same day. Dynamic Workflows hits Claude Code research preview — parallel hundreds of subagents handling repo-scale migrations. claude.ai and Cowork gain effort level controls — pick Claude's thinking depth, higher effort = better outcomes for faster rate-limit burn. The API now accepts system entries directly inside the messages array — let developers update instructions mid-agent-run without breaking prompt cache.
Next: Anthropic confirms it's building a new model class above Opus — the Mythos tier. Still in limited testing in cybersecurity. Expected to broaden to all users within weeks.
via Anthropic
