The model is live. Claude Opus 4.8 (API ID: claude-opus-4-8) is available across claude.ai, the API, and all integrations at the same price as Opus 4.7: $5 per million input tokens, $25 per million output tokens. Fast mode runs at 2.5x speed for $10/$50, which Anthropic says is three times cheaper than previous fast modes.
The benchmark picture matters more than any single number. Agentic coding (SWE-Bench Pro) jumps from 64.3% to 69.2%, beating GPT-5.5’s 58.6% and Gemini 3.1 Pro’s 54.2%. Multidisciplinary reasoning with tools (Humanity’s Last Exam) rises from 54.7% to 57.9%. Opus 4.8 is the first model to break 10% overall on the Legal Agent Benchmark’s all-pass standard, and the only model to complete every case end-to-end on the Super-Agent benchmark.
What’s structurally new is the self-correction behavior. Anthropic reports the model is four times less likely than Opus 4.7 to let flaws in its own code pass unremarked, and Bridgewater Associates specifically validated that it “proactively flags issues with the inputs and outputs of an analysis, something other models routinely missed.” That’s an enterprise reliability signal, not a marketing claim.
Comparison
Two platform features ship alongside the model. Dynamic workflows, a research preview in Claude Code, let the model generate a plan, spin up hundreds of parallel subagents to execute it, and verify results before reporting back. Anthropic says this enables “codebase-scale migrations across hundreds of thousands of lines of code from kickoff to merge.” Available for Enterprise, Team, and Max plans. Separately, effort control lets users on claude.ai and Cowork select how much thinking depth Claude applies, from Low (faster, lower rate-limit usage) to Max. Opus 4.8 defaults to High.
The 41-day release cycle is the signal underneath the benchmarks. Opus 4.7 launched in April to what TechCrunch described as a “chilly reception.” Opus 4.8 arriving this fast, with pointed improvements to self-correction and agentic reliability, suggests Anthropic is responding directly to competitive pressure from OpenAI’s Codex and Google’s Gemini Flash launches.
What to Watch
On the safety side, Anthropic reports alignment scores at “new highs on measures of prosocial traits” with misaligned behavior rates “substantially lower than Opus 4.7.” A full system card is published alongside the release. The Mythos preview, held since April for select organizations on cybersecurity work, is reportedly weeks away from broader release pending safeguards under Project Glasswing.
TJS Synthesis: Opus 4.8 isn’t a generational leap. It’s a targeted correction, fixing the self-awareness and reliability gaps that made Opus 4.7’s reception lukewarm, while adding the dynamic workflow infrastructure that enterprise buyers actually need. The compressed release cycle is the real story: Anthropic is shipping on a cadence that suggests it can’t afford to let a mid-tier release sit in market while OpenAI and Google iterate. For enterprise teams evaluating agentic AI platforms, the dynamic workflows feature and the Bridgewater validation are worth testing now. For everyone else, the Mythos timeline is the next decision point.