Anthropic has launched Claude Opus 4.8, a new AI model with improved benchmark performance and a greater tendency to flag uncertainties in its responses. The release includes new features like "Dynamic Workflows" for large-scale coding tasks and user-controlled effort settings. Anthropic also announced it expects to release its more powerful Mythos model, currently in a limited preview due to cybersecurity concerns, in the coming weeks.
Claude Opus 4.8 release
- ▪The pricing for regular usage of Claude Opus 4.8 is unchanged from its predecessor
- ▪Anthropic launched Claude Opus 4.8 on May 28, 2026, the newest version of its most advanced publicly available model
- ▪The release of Opus 4.8 came just 41 days after Opus 4.7, a faster upgrade cycle for Anthropic
- ▪The accelerated release schedule follows new model releases from competitors OpenAI and Google
Benchmark performance improvements
- ▪The model scored 84% on the Online-Mind2Web benchmark, a significant improvement over both Opus 4.7 and GPT-5.5
- ▪On the Super-Agent benchmark, Claude Opus 4.8 was the only model to complete every case end-to-end, outperforming previous Opus models and GPT-5.5
- ▪Claude Opus 4.8 shows improved performance on benchmarks for coding, agentic skills, reasoning, and practical knowledge work tasks
Enhanced honesty in responses
- ▪According to a testimonial from Bridgewater associates, the model proactively flags issues with the inputs and outputs of an analysis
- ▪Claude Opus 4.8 is approximately four times less likely than its predecessor to allow flaws in code it has written to pass unremarked
- ▪Early testers report that Claude Opus 4.8 is more likely to flag uncertainties and less likely to make unsupported claims
Dynamic workflows feature
- ▪Anthropic launched a new feature called "Dynamic Workflows" in research preview alongside the new model
- ▪Using this feature, Claude Code with Opus 4.8 can carry out codebase-scale migrations across hundreds of thousands of lines of code
- ▪Dynamic Workflows allows Claude Code to manage complex tasks by running hundreds of parallel subagents in a single session
Effort control settings
- ▪The "fast mode" for Opus 4.8 is now three times cheaper than it was for previous models
- ▪A new control allows users on claude.ai to choose how much effort the model puts into a response
- ▪Higher effort settings yield better responses by using more tokens, while lower settings are faster and use rate limits more slowly
Claude Mythos upcoming model
- ▪As part of Project Glasswing, a preview of Mythos is being used for cybersecurity work by companies including Amazon, Microsoft, and Apple
- ▪Anthropic plans to release its more advanced Mythos model to all customers in the "coming weeks."
- ▪The general release of the Mythos model has been held back due to its advanced cybersecurity capabilities, which have raised concerns
Story comments
Loading comments…