Published: 29 May 2026
Anthropic released Claude Opus 4.8 on 28 May 2026, the latest upgrade to its flagship AI model. Opus 4.8 improves on its predecessor across coding, agentic reasoning, and knowledge work benchmarks, while introducing a new “dynamic workflows” feature that allows Claude Code to run up to 1,000 parallel subagents in a single session. The model is available immediately at the same price as Opus 4.7: $5 per million input tokens and $25 per million output tokens.
What is Claude Opus 4.8 and what can it do?
Claude Opus 4.8 is Anthropic’s most capable generally available model, succeeding Claude Opus 4.7. It supports a 1 million token input context window with up to 128,000 output tokens, accepting both text and image inputs. Anthropic describes the model as having “sharper judgement” and greater honesty about its own progress, making it more reliable for extended agentic tasks.
Early testers report that Opus 4.8 is more likely to flag uncertainties in its work and less likely to make unsupported claims. According to Anthropic’s own evaluations, the model is around four times less likely than Opus 4.7 to let flaws in code it has written pass unremarked. Cursor’s CEO Michael Truell noted that Opus 4.8 “exceeds prior Opus models across every effort level” on CursorBench, with meaningfully more efficient tool calling.
Claude Opus 4.8 benchmarks and technical specifications
Claude Opus 4.8 posts measurable gains across multiple evaluation categories.
- On SWE-bench Verified, it scores 88.6%.
- Its Terminal-Bench 2.1 score is 74.6%.
- It reaches 1,890 Elo on GDPval-AA.
- Agentic coding performance increased from 64.3% (Opus 4.7) to 69.2%.
- Multidisciplinary reasoning with tools jumped from 54.7% to 57.9%.
- Agentic computer use moved from 82.8% to 83.4%.
- The knowledge work score rose from 1,753 to 1,890.
Anthropic states that Opus 4.8 beats GPT-5.5 across at least 12 benchmarks, including most knowledge-work, coding, agentic tool-use, and long-context evaluations. On alignment, Anthropic’s internal assessment concluded that Opus 4.8 “reaches new highs on our measures of prosocial traits” and shows rates of misaligned behavior substantially lower than Opus 4.7, comparable to Claude Mythos Preview.
Dynamic workflows: Claude Opus 4.8 can now run hundreds of parallel subagents
The most significant new feature launching alongside Opus 4.8 is dynamic workflows, available as a research preview in Claude Code. This feature lets Claude plan a large task, spawn hundreds of parallel subagents (capped at 1,000), and verify its own outputs before reporting back. It is designed for tasks too large for a single context window.
Anthropic gives codebase-scale migrations as the primary use case: Claude Code with Opus 4.8 can now carry out migrations across hundreds of thousands of lines of code from kickoff to merge, using the existing test suite as its quality bar. Dynamic workflows is available on Claude Code’s Enterprise, Team, and Max plans.
Additionally, Anthropic introduced effort control on claude.ai and Cowork. Users can now choose how much effort Claude puts into a response: lower settings give faster replies and slower rate-limit consumption, while higher settings produce better results. Opus 4.8 defaults to high effort. The Messages API also now accepts system entries inside the messages array, allowing developers to update instructions mid-task without breaking the prompt cache.
How does Claude Opus 4.8 compare to GPT-5.5 and Gemini 3.5 Flash?
Claude Opus 4.8 positions itself directly against OpenAI’s GPT-5.5 and Google’s Gemini 3.5 Flash. Anthropic claims Opus 4.8 outperforms GPT-5.5 on at least 12 benchmarks covering coding, knowledge work, agentic tool use, and long-context tasks. On SWE-bench Verified (88.6%) and Terminal-Bench 2.1 (74.6%), the scores place it among the top-performing models currently available.
At $5/$25 per million tokens (input/output), Opus 4.8 remains more expensive than Gemini 3.5 Flash ($1.50/$9.00) but is priced identically to its predecessor. The fast mode pricing has dropped significantly: $10/$50 per million tokens, down from $30/$150 for Opus 4.7, making it three times cheaper for speed-optimized workloads. Opus 4.8 fast mode runs at 2.5 times the standard speed.
Claude Opus 4.8 availability and pricing
Claude Opus 4.8 is available everywhere today. Standard pricing is $5 per million input tokens and $25 per million output tokens, unchanged from Opus 4.7. Fast mode costs $10 per million input tokens and $50 per million output tokens. Developers can access the model via the Claude API using the identifier claude-opus-4-8. The model is also available on AWS Bedrock as of launch day.
Anthropic also previewed what comes next: a new class of model with higher intelligence than Opus, building on Project Glasswing and Claude Mythos Preview. The company expects to bring Mythos-class models to all customers in the coming weeks, pending completion of additional cyber safeguards.
The full technical details and alignment assessment are available in the Claude Opus 4.8 System Card on Anthropic’s website.
