14 November 2026 · 2 min read
Anthropic Launches Claude Opus 4.8 With Sharper Coding and a Focus on Honesty
Claude Opus 4.8 brings better agentic coding, a 4x reduction in silent code errors, and new developer controls, while running faster and cheaper than its predecessor.
Anthropic has launched Claude Opus 4.8, a model that the company says redefines how developers handle programming and complex analytical tasks. The update puts a strong emphasis on agentic capabilities and, unusually, on admitting uncertainty rather than hallucinating answers [1].
What is new in Opus 4.8
Anthropic describes Opus 4.8 as a “more effective collaborative partner” for agentic tasks. It can plan, analyse interdisciplinary problems, and control computer interfaces to complete integrated work with higher efficiency than previous versions [1].
Early testers noted that the model is not only smarter and sharper in decision-making but also more honest. It is now more inclined to flag uncertainty instead of inventing baseless claims. In practical terms, Opus 4.8 is four times less likely to pass along code containing errors without flagging them compared to earlier releases [1].
Benchmark performance
On the SWE-Bench Pro programming test, Opus 4.8 scored 69.2%, placing it ahead of OpenAI’s GPT-5.5 and Google’s Gemini 3.1 Pro. However, GPT-5.5 still holds an edge in terminal-coding tests [1].
Speed and cost
Fast mode in Opus 4.8 operates 2.5 times faster than previous versions and is three times cheaper. Regular usage prices remain unchanged from Opus 4.7 [1].
New developer controls
Anthropic introduced several features alongside the model:
- Dynamic Workflows (beta for researchers and Business, Team, and Max plans): lets the model plan large tasks and run hundreds of sub-agents in parallel within a single session, enabling management of massive software projects [1].
- Effort Control: users on Claude.ai and Cowork can choose how much effort the model puts into an answer. Lower effort gives faster responses and consumes quota more slowly. Opus 4.8 defaults to High Effort mode [1].
- Messages API updates: developers can now update core instructions mid-task, adding flexibility that was not previously available [1].
Safety
Evaluations indicate that Opus 4.8 reaches new record levels in positive social traits, such as respecting user autonomy and acting in the user’s best interest, with a noticeable decrease in misleading or deceptive behaviours [1].
Availability
Claude Opus 4.8 launched worldwide immediately, with regular pricing unchanged from Opus 4.7. The faster and cheaper fast mode is also available [1].
What it means for developers
For builders, Opus 4.8 offers a more reliable coding partner that explicitly flags uncertainty instead of silently producing broken code. The combination of stronger agentic planning, faster and cheaper inference, and new controls over effort and mid-task instructions makes it a practical upgrade for complex software projects. The Dynamic Workflows feature, even in beta, points toward handling large-scale codebases with parallel sub-agents, which could change how teams approach refactoring and migration work.
Anthropic also confirmed it is working on models with similar capabilities at lower cost and on an entirely new class of models that exceed Opus’s intelligence. The company is putting finishing touches on the Claude Mythos model, expected to release to all customers within a few weeks [1].
Sources
- anthropic
- claude
- opus-4-8
- ai-coding
- llm
Researched and drafted with AI from the sources above.
Learn to build with AI
Hands-on tutorials on Cursor, Claude Code, AI agents and more.