Anthropic launches Claude Opus 5 with half the cost of Fable 5 and stronger coding performance
Anthropic has introduced Claude Opus 5, the latest version of its flagship AI model. The company says the new model delivers performance close to Claude Fable 5 while costing half as much, making it a more practical choice for everyday use. Opus 5 arrives just two months after Opus 4.8 and follows the launch of Mythos 5, Fable 5 and Sonnet 5 in June. With this release, Haiku remains the only Claude model yet to receive a 5-series upgrade. 
Also Read: OpenAI says AI model escaped testing and breached Hugging Face during security test
Anthropic says Opus 5 is now the default model for Claude Max users and the most capable option available on the Claude Pro plan. It is also available through the Claude API for developers. Also Read: Google Meet users no longer have to search for recordings thanks to this new update
Better performance without increasing the price
According to Anthropic, Opus 5 offers better performance than Opus 4.8 while keeping the same pricing of $5 per million input tokens and $25 per million output tokens. The company says developers can also choose different effort settings depending on whether they want higher accuracy or lower costs. Also Read: Meta tests StoryKit, an AI app that creates bedtime stories in seconds
Anthropic claims Opus 5 performs particularly well on coding and knowledge-based tasks. On Frontier-Bench v0.1, it outperformed other models while completing tasks at a lower cost than Opus 4.8. On CursorBench 3.2, the model came within 0.5 per cent of Claude Fable 5’s best score while costing roughly half as much per task.
The company also reported strong results on other benchmarks, including ARC-AGI 3, AutomationBench, OSWorld 2.0, GDPval-AA v2, HLE and DeepSearchQA. Anthropic says the model also performs better than Opus 4.8 on scientific research tasks covering structural biology, organic chemistry and bioinformatics.
Improved reasoning and coding
Anthropic says one of the biggest changes in Opus 5 is how it checks its own work before responding. According to the company, the model is better at verifying answers, identifying mistakes and refining its approach until it completes a task successfully.
As an example, Anthropic said Opus 5 was given an incomplete prompt asking it to recreate a machine part without being able to directly view the drawing. The model reportedly built its own computer vision pipeline to extract the required geometry before successfully generating the 3D model. The company also said Opus 5 identified the root cause of a real software bug that another model failed to fix completely.
Fewer restrictions than Fable 5
Alongside the performance improvements, Anthropic has also reduced some of the safety restrictions that users experienced with Claude Fable 5. Like Opus 4.8, the new model does not come with the 30-day data retention policy that applies to Fable 5 and Mythos 5, which had raised concerns among privacy-focused users.
The company says Opus 5 still includes safeguards for cybersecurity-related requests. It can review source code to identify software vulnerabilities, but it blocks tasks such as binary-based vulnerability scanning, penetration testing and exploit generation. Anthropic expects these safety classifiers to intervene around 85 per cent less often than they do with Fable 5.
Anthropic has also introduced a beta feature called Automatic Fallbacks. Instead of blocking a request when safety systems are triggered, the feature automatically routes it to another compatible model, allowing users to receive a response instead of an error message. Fast mode is also available, running around 2.5 times faster than the default version at a higher usage cost.
