Overview
Claude Opus 4.8 is Anthropic's May 28, 2026 Opus-tier flagship — an upgrade over Opus 4.7 that improves across benchmarks while holding the same $5/$25 per-million-token pricing. Anthropic frames it as "a more effective collaborator": better honesty and judgment, stronger agentic decision-making, and more efficient tool use, rather than a single headline capability jump. Opus 4.8 remained Anthropic's most capable model until Claude Fable 5 shipped on June 9, 2026 — and it is still the model Anthropic recommends for most complex Opus-tier work, and the launch fallback target when Fable 5 declines a request.
The model is available on claude.ai, the Claude API as claude-opus-4-8, Claude Code, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry (where the context window is 200k rather than 1M). It carries a 1M-token context window and up to 128k output tokens by default, with the effort parameter defaulting to high on all surfaces.
What's New
Stronger Honesty and Judgment
The headline improvement is reliability of judgment. Opus 4.8 is roughly 4× less likely than Opus 4.7 to let code flaws pass unremarked — a direct win for code review and agentic engineering, where a model that silently approves a broken change is worse than one that flags uncertainty. It is also the first model to break the 10% all-pass standard on the Legal Agent Benchmark, a demanding measure of end-to-end correctness on professional legal tasks.
More Capable Agents
Opus 4.8 makes better decisions and catches more of its own errors across long agentic tasks, and it calls tools more efficiently — reaching equivalent results in fewer steps. For teams running multi-step agents, fewer wasted tool calls means lower latency and cost per completed task on top of the quality gain.
Stronger Computer Use
On Online-Mind2Web, an evaluation of real-world browser and computer-use tasks, Opus 4.8 scores 84% — a meaningful improvement over Opus 4.7 and a competitive result against other frontier models. This makes it a stronger backbone for browser agents and screen-driven automation.
Better Long-Session Consistency
Opus 4.8 retains context and holds a consistent style across extended sessions more reliably than Opus 4.7. Combined with CursorBench results that exceed prior Opus models across every effort level, it is positioned as a steadier choice for long-running coding and knowledge work.
Performance Benchmarks
Headline numbers from Anthropic's release (vs. Opus 4.7 unless noted):
| Benchmark / Task | Claude Opus 4.8 | Baseline |
|---|---|---|
| Code-flaw oversight | ~4× less likely to miss | vs. Opus 4.7 |
| Legal Agent Benchmark | First to break 10% all-pass | — |
| Online-Mind2Web (computer use) | 84% | Improvement over Opus 4.7 and GPT-5.5 |
| CursorBench | Exceeds prior Opus at every effort level | Prior Opus models |
The honesty and computer-use gains matter most for teams running autonomous AI agent and AI code checker pipelines, where a model that flags rather than hides problems changes how much it can be trusted unattended.
Migration Guide
Opus 4.8 is a drop-in upgrade from Opus 4.7 at the same price, and it preserves Opus 4.7 platform compatibility. A few notes when moving over:
- Same tokenizer as Opus 4.7. Token counts are consistent with Opus 4.7, so context budgets and cost estimates carry over without the re-estimation needed when 4.7 first changed the tokenizer.
- Effort defaults to
high. On all surfaces, including the Claude API and Claude Code,effortdefaults tohigh. Set it explicitly if you need a different latency/quality point. - Adaptive thinking, not extended thinking. Opus 4.8 uses adaptive thinking; the older extended-thinking mode is not used on this model.
Pricing & Plans
| Plan | Price |
|---|---|
| API – Input (regular) | $5 per million tokens |
| API – Output (regular) | $25 per million tokens |
| API – Fast mode | $10 input / $50 output per million tokens |
| Claude.ai Pro / Max | Included within plan limits |
| Team / Enterprise | Contact Anthropic for volume pricing |
| Cloud availability | Amazon Bedrock, Google Cloud Vertex AI, Microsoft Foundry (200k context) |
Fast mode note: Opus 4.8's optional fast mode doubles the per-token price to $10/$50 in exchange for faster output. Standard requests stay at $5/$25 — the same as Opus 4.7.
Best For
- Engineering teams running code review and autonomous coding agents that need a model which flags problems instead of silently approving them
- AI agent and browser-automation platforms that depend on strong computer-use performance
- Legal, finance, and professional-services teams needing end-to-end correctness on multi-step agentic tasks
- Developers on Claude Code who want better long-session consistency at Opus 4.7 pricing
- Teams that need an Opus-tier model below the cost of Fable 5 — including as the fallback target when Fable 5 declines a request
- Anyone on Sonnet 4.6 who needs Opus-tier judgment on their hardest workloads
FAQ
How is Claude Opus 4.8 different from Opus 4.7?
Opus 4.8 is an across-the-board upgrade at the same $5/$25 pricing. The biggest gains are in honesty and judgment (roughly 4× less likely to let code flaws pass), agentic decision-making and error detection, tool-calling efficiency, computer use (84% on Online-Mind2Web), and long-session consistency. It uses the same tokenizer as Opus 4.7, so costs and context budgets carry over.
Is Opus 4.8 still Anthropic's most capable model?
No. Opus 4.8 was the flagship until Claude Fable 5 launched on June 9, 2026. Opus 4.8 remains Anthropic's recommended Opus-tier model for most complex work and is the launch fallback target when Fable 5's safety classifiers decline a request.
How much does Claude Opus 4.8 cost?
Standard pricing is $5 per million input tokens and $25 per million output tokens — the same as Opus 4.7. An optional fast mode delivers faster output at $10 input / $50 output per million tokens. On claude.ai it is included within Pro, Max, Team, and Enterprise plan limits.
Where can I use Claude Opus 4.8?
Opus 4.8 is available on claude.ai, the Claude API as claude-opus-4-8, Claude Code, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry. On Microsoft Foundry the context window is 200k tokens rather than the 1M available on other surfaces.
Should I upgrade from Opus 4.7 to Opus 4.8?
For most teams, yes — it is the same price and tokenizer with measurable gains in honesty, agentic reliability, and computer use, and it preserves Opus 4.7 platform compatibility. Test high-volume prompts on a sample first, but the migration is low-friction compared with earlier Opus transitions.
