Claude Opus 4.8 Review: What's New, Benchmarks & Pricing
Anthropic released Claude Opus 4.8 on May 28, 2026 — an incremental but meaningful upgrade to the Opus class. Here's what changed, what the benchmarks show, and whether it's worth switching from Opus 4.7.
Reviewed 2026-06-10 · Source: Anthropic announcement
Quick Verdict
Claude Opus 4.8 is a solid upgrade — especially for developers running agentic workflows. The honesty improvements (4× fewer false code-confidence claims) and computer use gains (84% on Online-Mind2Web) are the most practically valuable changes. Pricing stays the same as Opus 4.7 ($5/$25 per million tokens), so there's no cost reason to stay on 4.7. Anthropic itself calls the improvement "modest but tangible," which is honest — this is not a generational leap.
What's New in Opus 4.8
Dynamic workflows in Claude Code
Claude can now plan a large task and spin up hundreds of parallel subagents in a single Claude Code session, then verify outputs before reporting back. Anthropic demonstrated codebase-scale migrations across hundreds of thousands of lines of code — from kickoff to merge — using the existing test suite as the quality bar. Available on Enterprise, Team, and Max plans.
Effort control
A new slider alongside the model selector in claude.ai and the API lets you choose how much compute Opus 4.8 uses per response. Levels: low, high (default), extra, and max. Higher effort produces better results but uses more tokens and is slower. Anthropic also raised rate limits in Claude Code to accommodate higher-effort workloads.
System entries mid-task via Messages API
Developers can now insert system entries inside the messages array, allowing instructions to be updated mid-agent-run without breaking the prompt cache or routing the update through a user turn. Useful for changing permissions, token budgets, or environment context as an agent progresses.
Honesty improvements
Opus 4.8 is trained to flag uncertainty more aggressively. Evals show it is around 4× less likely than Opus 4.7 to allow flaws in code it has written to pass unremarked — a significant change for agentic coding workflows where undetected errors compound across steps.
Benchmark Comparison: Opus 4.8 vs 4.7
Key evaluation results from Anthropic's release and third-party partners.
| Benchmark | Opus 4.8 | Opus 4.7 |
|---|---|---|
| Online-Mind2Web (computer use) | 84% | 82.3% |
| Legal Agent Benchmark (all-pass) | >10% | <10% |
| False code-confidence rate | 4× lower than Opus 4.7 | baseline |
| Token cost vs Opus 4.7 (multimodal retrieval) | 61% cheaper | baseline |
Pricing
Unchanged from Opus 4.7. Same price, better model.
Input: $5 / million tokens
Output: $25 / million tokens
Same as Opus 4.7. API access via claude-opus-4-8.
Input: $10 / million tokens
Output: $50 / million tokens
Higher throughput, lower latency. Doubles the token cost.
API model ID: claude-opus-4-8. Available via Anthropic Console.
Who Should Upgrade to Opus 4.8?
Upgrade now if you're running:
- ✓ Agentic coding pipelines where false confidence causes compounding errors
- ✓ Computer use or browser automation workflows
- ✓ Long-running multi-step professional tasks (legal, finance)
- ✓ High-context retrieval over PDFs and documents (61% cheaper)
Stay on 4.7 if:
- → You're budget-sensitive and Sonnet 4.6 already covers your use case at lower cost
- → Your evals show Opus 4.7 output quality is already satisfactory
- → You haven't yet tested effort control — the default "high" mode uses more tokens than 4.7
Also released alongside Opus 4.8
- • Dynamic workflows in Claude Code — hundreds of parallel subagents per session (Enterprise/Team/Max)
- • Effort control in claude.ai and API — tune quality vs. token usage per request
- • Messages API system entries — update Claude's instructions mid-task without breaking prompt cache
- • Project Glasswing preview — Claude Mythos models available to select cybersecurity organizations; general release coming soon
Frequently Asked Questions
What is Claude Opus 4.8?
Claude Opus 4.8 is Anthropic's flagship large language model, released May 28, 2026. It is an incremental upgrade to Opus 4.7 — Anthropic describes improvements as 'modest but tangible.' The headline changes are: better honesty (4× reduction in false code-confidence), higher computer use benchmark scores (84% on Online-Mind2Web), improved efficiency for multimodal retrieval tasks, and new infrastructure features including dynamic workflows and effort control.
How does Claude Opus 4.8 compare to Opus 4.7?
Opus 4.8 improves over 4.7 across coding reliability, computer use, agentic tasks, and legal/professional benchmarks. The most notable change is honesty: Opus 4.8 is far less likely to claim it has completed a task correctly when it hasn't. Computer use improved from ~82% to 84% on Online-Mind2Web. Token efficiency for multimodal retrieval is reported as 61% lower cost by Databricks. Pricing is unchanged.
What is Claude Opus 4.8 pricing?
Pricing is unchanged from Opus 4.7: $5 per million input tokens and $25 per million output tokens for standard usage. Fast mode pricing is $10 per million input tokens and $50 per million output tokens. Anthropic raised Claude Code rate limits to accommodate the higher token consumption of high-effort mode.
What are dynamic workflows in Claude Code?
Dynamic workflows, available in research preview with Opus 4.8, let Claude plan a large engineering task and then launch hundreds of parallel subagents within a single Claude Code session. The orchestrating Claude instance verifies outputs before reporting completion. Anthropic demonstrated full codebase migrations using this feature. It is available on Claude Code Enterprise, Team, and Max plans.
What is effort control in Claude Opus 4.8?
Effort control is a new setting available in claude.ai and the API that lets you choose how much compute Opus 4.8 applies per response. The options are low, high (default), extra, and max. Higher effort yields better results at higher token cost and slower latency. In Claude Code, the equivalent is 'xhigh.' Anthropic recommends 'extra' for difficult tasks and long async workflows.
Is Claude Opus 4.8 available via API?
Yes. Developers can access it via the Anthropic API using the model ID claude-opus-4-8. The model supports tool use, vision, extended context (200K tokens), and the new mid-task system message feature. All existing Claude API integrations can switch to Opus 4.8 by updating the model string.
Try Claude Opus 4.8
Available today on claude.ai (Pro and above) and via the Anthropic API.
Try Claude →API developers: model ID is claude-opus-4-8
Affiliate disclosure: Some links on this page are affiliate links. If you sign up through them, AISO Tools may earn a commission at no extra cost to you. This never affects our rankings or reviews.
📬 Get the best new AI tools delivered weekly
One concise email with fresh launches, trending picks, and featured standouts.
Join thousands of professionals who discover the best AI tools every week. No spam — unsubscribe anytime.