Following significant advancements from competitors like OpenAI’s GPT-5.5 and Google’s Gemini 3.5 Flash, Anthropic has launched Claude Opus with version 4.8. This update to Opus 4.7 maintains the same base pricing while delivering substantial improvements in benchmarks, collaboration, agentic tasks, reasoning, programming, and long-form work with tools.
Anthropic has also introduced several practical new features, including a more affordable fast mode, an effort control to adjust quality or speed, and dynamic workflows for Claude Code.
Opus 4.8: The Fastest AI Ever Created Outperforms the Competition
If GPT-5.5 represented a significant leap forward, particularly in reasoning, Opus 4.8 has now surpassed it in the ongoing competition between leading AI companies, alongside Google.
The headline figures are striking: Opus 4.8 can operate 2.5 times faster in fast mode. This accelerated mode is now 3 times cheaper than in previous versions. Anthropic also claims the model is approximately 4 times less likely than Opus 4.7 to overlook hidden bugs in its own generated code. This is particularly significant for developers, as Anthropic appears to have closed the gap with OpenAI’s Codex, at least on paper.
This improvement in programming is crucial, not just for generating more code, but for accurately identifying and flagging errors before presenting the work as perfect.
Regarding pricing, Anthropic has kept the standard cost for Opus 4.8 at $5 per million input tokens and $25 per million output tokens. The fast mode, however, is priced at $10 per million input tokens and $50 per million output tokens, but in return offers the 2.5 times superior speed. While still a premium, it’s notably more cost-effective than with version 4.7.
Hundreds of Agents at Your Disposal
Claude Code sees its most significant advancement with dynamic workflows. This preview feature allows for the division of enormous tasks, their planning, and the execution of hundreds of sub-agents in parallel within a single session. Anthropic cites large-scale codebase migrations spanning hundreds of thousands of lines of code as an example, from initial work through to merging, using existing test suites for validation.
Publicly released metrics from the presentation show Opus 4.8 achieving 69.2% on SWE-Bench Pro, compared to Opus 4.7’s 64.3%, GPT-5.5’s 58.6%, and Gemini 3.1 Pro’s 54.2%. It scores 74.6% on Terminal-Bench 2.1 and reaches 83.4% on OSWorld-Verified. Additionally, it achieved 1,890 Elo points in GDPval-AA, a 53.9% score on Finance Agent v2, and 57.9% on Humanity’s Last Exam when using tools.
Anthropic also shared partner data: 84% on Online-Mind2Web for browser and computer usage, the first result exceeding 10% globally in a legal benchmark with an all-pass standard, and up to 61% lower token cost compared to Opus 4.7 in multimodal workflows involving PDFs, diagrams, and unstructured content. The overarching theme is enhanced performance, reduced waste, and greater capability for long-form tasks where minor errors can be critical. Therefore, Opus 4.8 represents an impressive incremental improvement, arriving just before the anticipated launch of Mythos, which promises to be a game-changer.
