Claude Opus 4.8: Pricing, Fast Mode, Workflows

What changed in Claude Opus 4.8 compared with Opus 4.7

Claude Opus 4.8 is a same-price upgrade to Opus 4.7 from Anthropic, the maker of Claude. The company calls it a modest but tangible improvement, with gains across its benchmarks. Three additions launched with it: an effort control in claude.ai and Cowork, dynamic workflows in Claude Code, and a cheaper fast mode.

The effort control sets how hard Claude works on a response. Dynamic workflows are a research preview for very large jobs in Claude Code. Fast mode runs the model at 2.5 times the speed.

Claude Opus 4.8 release date, API model ID and availability

Anthropic announced Claude Opus 4.8 on May 28, 2026, and said it was available everywhere that day. Developers who call it through the Claude API use the model ID claude-opus-4-8.

The announcement does not list which chat plans include the model beyond that. It names plans only for dynamic workflows.

Claude Opus 4.8 pricing: standard and fast mode cost

Standard use of Claude Opus 4.8 costs $5 for input and $25 for output per million tokens, the same as Opus 4.7. A token is a small chunk of text, and prices are quoted per million of them. Opus 4.6, 4.7 and 4.8 are compared line by line in "Claude Opus 4.6 vs 4.7 vs 4.8 pricing compared".

Fast mode is a separate tier that returns answers sooner. Input costs $10 and output costs $50 per million tokens, twice the standard rate, while the model runs at 2.5 times the speed. Anthropic says fast mode is now three times cheaper than it was for earlier models.

Dynamic workflows: Claude Opus 4.8 in Claude Code

Dynamic workflows let Claude Code split a very large job across many helpers. Claude drafts a plan, then runs hundreds of subagents in parallel inside one session, each a separate instance working on part of the job. Before reporting back, Claude verifies the results, and with Opus 4.8 the agents can keep running longer.

Anthropic's example is a codebase migration: Claude Code with Opus 4.8 can carry one across hundreds of thousands of lines of code, from kickoff to merge, using the project's existing test suite as the bar. The example assumes a large codebase and an existing test suite to check the migrated code against.

The feature is a research preview. It is available in Claude Code for Enterprise, Team and Max plans.

Effort control and the default effort in Claude Opus 4.8

Opus 4.8 defaults to high effort, and claude.ai and Cowork now show an effort control next to the model selector on all plans. Higher settings make Claude think more often and more deeply. Lower settings make it answer faster and use up rate limits more slowly.

Anthropic judges high effort the best balance of quality and user experience. On coding tasks it spends about as many tokens as Opus 4.7's default did, with better results.

Two higher levels exist: "extra" (called xhigh in Claude Code) and "max", where the model spends more tokens for better results. Anthropic recommends "extra" for difficult tasks and long-running asynchronous workflows. It also raised Claude Code rate limits to cover the heavier token use.

Claude Opus 4.8 benchmarks, honesty and the system card

Honesty is one of the most prominent improvements Anthropic cites for Opus 4.8: in its tests, the model lets flaws in its own code pass without comment around four times less often than Opus 4.7. Early testers also report that it flags uncertainty more often and makes fewer unsupported claims.

The alignment team found that misaligned behavior, such as deception or cooperation with misuse, is substantially lower than in Opus 4.7. The rate is similar to Claude Mythos Preview, which Anthropic calls its best-aligned model. The full assessment and the pre-deployment safety tests are in the system card.

A tester quoted in the announcement reports 84% on Online-Mind2Web, ahead of both Opus 4.7 and GPT-5.5. That figure comes from the tester, not from Anthropic's own tests.

The comparison table with Anthropic's scores is on the announcement page, and two footnotes affect how to read it. Terminal-Bench 2.1 scores use the Terminus-2 public harness, while GPT-5.5 also has a separate score reported with the Codex CLI harness. The OSWorld-Verified method changed, so the Opus 4.7 score was revised to 82.3%.

Claude Opus 4.8 API change: system entries in messages

The Messages API now accepts system entries inside the messages array. Before this, changing Claude's instructions partway through a task meant either sending the change as a user turn or breaking the prompt cache. Now a developer can add the update in place.

The update does not break the prompt cache, the store that lets repeated input be reused instead of processed again. Anthropic suggests using it to change permissions, token budgets or environment context while an agent runs.

Claude Opus 4.8 and Mythos: what Anthropic said is next

As of May 28, 2026, Anthropic named two next steps: models that offer many of Opus's capabilities at a lower cost, and a new class of model with higher intelligence than Opus. The second class is Claude Mythos.

A small number of organizations use Claude Mythos Preview for cybersecurity work under Project Glasswing. Anthropic said it expected Mythos-class models to reach all customers within weeks. That was a forecast made on the launch date, not a confirmed release.

Claude Opus 4.8 summary: price, access and what to try

Claude Opus 4.8 launched on May 28, 2026 as a same-price upgrade to Opus 4.7. Standard use costs $5 and $25 per million tokens, and fast mode costs $10 and $50. Here is where each change applies:

• Model: the API model ID is claude-opus-4-8.
• Effort control: next to the model selector in claude.ai and Cowork, on all plans.
• Dynamic workflows: Claude Code on Enterprise, Team and Max plans, as a research preview.
• Messages API: system entries are now allowed inside the messages array.
• Full test results: the Claude Opus 4.8 system card.