Anthropic Debuts Claude Opus 4.8

Anthropic has released Claude Opus 4.8, an upgraded version of its flagship AI model. The update promises enhanced performance across various benchmarks and improved collaborative capabilities.

The new Claude Opus 4.8 builds upon the previous Opus 4.7 iteration, showcasing advancements in areas like coding, agentic skills, and reasoning. Early testers report increased reliability and sharper judgment in agentic tasks.

New Features and Performance Gains

Claude Opus 4.8 introduces several new features. Users on claude.ai can now adjust the computational 'effort' Claude expends on a task, balancing response speed with depth. Claude Code gains 'dynamic workflows,' enabling it to handle extremely large-scale coding problems, such as complete codebase migrations. The fast mode for Opus 4.8 is now three times cheaper than for previous models, operating at 2.5x speed.

On benchmarks, Opus 4.8 demonstrates significant improvements. For instance, it's the only model to complete all cases end-to-end on the Super-Agent benchmark, surpassing prior Opus models and GPT-5.5. It also achieved the highest score on the Legal Agent Benchmark, offering a substantial accuracy lift for legal professionals.

Early adopter feedback highlights Opus 4.8's enhanced reliability and judgment. Tom Pritchard, Staff Engineer, noted its improved ability to ask pertinent questions, self-correct, and push back on unsound plans within Claude Code. Kay Zhu, Co-Founder and CTO, confirmed its superior performance on the Super-Agent benchmark, outperforming previous Opus versions and matching GPT-5.5 in cost-effectiveness for agent products.

Michael Truell, Co-Founder and CEO, pointed to significant efficiency gains in tool calling on CursorBench, with Opus 4.8 using fewer steps for comparable intelligence. Niko Grupen, Head of Applied Research, lauded its performance on the Legal Agent Benchmark, indicating a direct translation to reduced attorney workload for customers.

Katie Parrott, Staff Writer, described Opus 4.8 as a major quality-of-life improvement over Opus 4.7, being faster, easier to collaborate with, and better at maintaining context and style. Miguel Gonzalez, Tech Lead, called it the strongest computer-use and browser-agent model tested, scoring 84% on Online-Mind2Web, a notable increase over Opus 4.7 and GPT-5.5.

Scott Wu, CEO, emphasized Opus 4.8's clean tool usage and instruction following, crucial for autonomous engineering workloads. Joel Hron, Chief Technology Officer, highlighted consistency and reasoning quality improvements in CoCounsel Legal, raising the bar for trusted AI performance.

Hanlin Tang, CTO at Neural Networks, stated that Opus 4.8 unlocks a step change in agentic reasoning for Databricks' Genie, handling complex, multi-step questions faster and at a 61% lower token cost for PDF and diagram reasoning. Aabhas Sharma, CTO at Hebbia, noted improved citation precision and token efficiency for financial-document workflows.

Enhanced Honesty and Alignment

A key improvement in Opus 4.8 is its increased honesty, with testers reporting it's more likely to flag uncertainties and less prone to making unsupported claims. Evaluations show it's four times less likely than its predecessor to overlook flaws in generated code.

Anthropic's alignment assessment confirmed that Opus 4.8 reaches new highs in prosocial traits, supporting user autonomy and acting in the user's best interest. Misaligned behaviors, such as deception or cooperation with misuse, are substantially lower than Opus 4.7, nearing the performance of their best-aligned model, Claude Mythos Preview.

Broader Availability and Future Outlook

Beyond Opus 4.8, Anthropic is rolling out 'dynamic workflows' for Claude Code, allowing for hundreds of parallel subagents to tackle massive tasks. Effort control is now available on claude.ai and Cowork, letting users fine-tune response quality and speed. The Messages API also supports system entries within the messages array for more dynamic instruction updates.

Anthropic plans to release models offering similar capabilities at lower costs and is developing a new class of models with even higher intelligence, including Mythos-class models for cybersecurity work, expected to be generally available soon.

Claude Opus 4.8 is available today at the same pricing as Opus 4.7: $5 per million input tokens and $25 per million output tokens. Fast mode pricing is $10 per million input tokens and $50 per million output tokens. Developers can access it via the Claude API using claude-opus-4-8.