Anthropic has launched Claude Opus 4.8, an Opus model focused on better coding, stronger agentic work, effort controls, and lower fast mode pricing. It is available today, and developers can use it through the Claude API with the model ID claude-opus-4-8.
Regular API pricing stays unchanged from Opus 4.7 at $5 per million input tokens and $25 per million output tokens. Fast mode now costs $10 per million input tokens and $50 per million output tokens, down from $30 and $150 for Opus 4.6 and Opus 4.7 fast mode.
| Benchmark | Opus 4.8 | Opus 4.7 | GPT-5.5 | Gemini 3.1 Pro |
|---|---|---|---|---|
| SWE-Bench Pro | 69.2% | 64.3% | 58.6% | 54.2% |
| Terminal-Bench 2.1 | 74.6% | 66.1% | 78.2% | 70.3% |
| Humanity’s Last Exam, no tools | 49.8% | 46.9% | 41.4% | 44.4% |
| Humanity’s Last Exam, with tools | 57.9% | 54.7% | 52.2% | 51.4% |
| OSWorld-Verified | 83.4% | 82.8% | 78.7% | 76.2% |
| GDPval-AA | 1890 | 1753 | 1769 | 1314 |
| Finance Agent v2 | 53.9% | 51.5% | 51.8% | 43.0% |
Source: Anthropic
What Opus 4.8 Changes
Opus 4.8 is not a dramatic reset. It is a cleaner update to the parts of Claude that matter in longer work: coding, agentic tasks, tool use, reasoning, and reliability.
Anthropic says the model builds on Opus 4.7 with improvements across benchmarks and collaboration. Opus 4.7 was already strong, but Opus 4.8 is being pushed harder for work where the model has to stay useful across more steps and more decisions.
That is also where the market is moving. OpenAI is pushing GPT-5.5 around coding, research, data analysis, and longer professional workflows. Google is pushing Gemini models around agentic and coding work. Opus 4.8 lands in that same fight, but with Anthropic’s usual angle: control, caution, and steadier collaboration.
Where Opus 4.7 Lagged
The most useful Opus 4.8 upgrade may be honesty. Anthropic says the new model is around four times less likely than Opus 4.7 to let flaws in its own code pass without comment.
That is not a promise that Claude will catch every mistake. It is still an evaluation claim. But it targets a real developer pain: AI code that sounds confident, but still hides a bad assumption.
New Effort Controls
Claude users now get an effort control in claude.ai. It lets users choose how much effort Claude puts into a task.
Higher effort is meant for harder work where deeper thinking helps. Lower effort is meant for faster replies and slower rate-limit use. Anthropic says the control is available on all plans, and Opus 4.8 defaults to high effort.
A quick rewrite should not use the same model effort as a difficult debugging session. This gives users a cleaner way to choose between speed and depth.
Dynamic Workflows in Claude Code
Claude Code is also getting dynamic workflows in research preview. Claude can plan larger engineering tasks, run parallel subagents, verify results, and report back.
Anthropic says dynamic workflows are available in Claude Code for Max, Team, and Enterprise users, with Enterprise access requiring admin enablement. It is better framed as a heavy-duty mode, since Anthropic says it can use far more tokens than a normal Claude Code session.
Why It Matters
Claude Opus 4.8 gives developers a new model ID, cheaper fast mode, stronger honesty claims, and a bigger-task workflow path inside Claude Code.
Opus 4.8 does not prove Claude is better than every other frontier model. It does make Claude more interesting for people who care about coding, agentic workflows, and models that can question their own work before the user has to.