Anthropic Introduces Claude Sonnet 5 With Enhanced Agentic Capabilities; Confirms Return of Claude Mythos 5, Fable 5

Anthropic has released its latest Claude Sonnet 5 AI model. The new AI model is claimed to bring enhanced agentic capabilities compared to the Claude Sonnet 4.6 model. Moreover, the new AI model is claimed to offer agentic search capabilities at par with the Opus 4.8 model. Along with improved performance, the AI model focuses on cost efficiency. It is claimed to offer similar performance as more capable AI models at lower costs. Separately, the AI giant has announced that the US government has lifted export restrictions on its Claude Mythos 5 and Mythos-class Fable 5 AI models. This comes weeks after the company was asked to restrict the availability of the two AI models for foreign nationals.

Claude Sonnet 5 Features, Capabilities

On Tuesday, the US-based AI giant released its latest Claude Sonnet 5 AI model for everyone globally. It is available at an introductory price of $2 (roughly Rs. 191) per 10,00,000 input tokens and $10 (about Rs. 952) per 10,00,000 output tokens till August 31. The standard price is set at $3 (roughly Rs. 286) and $15 (about Rs. 1,428), respectively. To accommodate the higher token usage, Anthropic has also increased the rate limits on Chat, Cowork, Claude Code, and Claude Platform.

In terms of capabilities, Anthropic claims that Claude Sonnet 5 delivers enhanced agentic search performance over the Sonnet 4.6 AI model. Meanwhile, it offers other agentic capabilities on par with the Opus 4.8 model, at lower costs. In terms of agentic coding, Sonnet 5 scored 63.2 percent on SWE-bench Pro, higher than Sonnet 4.6's 58.1 percent score, but lower than Opus 4.8's 69.2 percent score.

On Humanity's Last Exam's multidisciplinary reasoning test, the new Claude Sonnet 5 scored 57.4 percent with tools and 43.2 percent without tools, close to Opus 4.8's 57.9 percent and 49.8 percent scores, respectively. Citing BrowseComp's agentic search performance by effort level results, Claude Sonnet 5 managed to offer a pass rate percentage on par with Opus 4.8, at a lower cost per task, while achieving a significantly higher pass rate percentage than the Sonnet 4.6 AI model.

Anthropic says that testers found the Claude Sonnet 5 AI model much better at performing agentic tasks compared to older Sonnet models. During early tests, the AI model was also able to check its own output without being asked to do so. On top of this, the AI giant claims that Claude Sonnet 5 is better at refusing malicious requests, while also resisting hijack attempts in prompt injection attacks. It also presented lower rates of hallucinations and sycophancy than the Sonnet 4.6 AI model.

Claude Mythos 5, Fable 5 Global Redeployment Announced

Separately, Anthropic announced on Tuesday that the US government has removed export controls on the company's Claude Mythos 5 and Fable 5 AI models. This comes weeks after the US administration restricted the AI giant from making its most capable cybersecurity-trained Mythos 5 and the “Mythos-class” Fable 5 models available to foreign nationals.

The company said that Fable 5 is now available to users globally on the Claude Platform, Claude.ai, Claude Code, and Claude Cowork. Anthropic has also increased the weekly usage limits by 50 percent till July 7 for Pro, Max, Team, and select Enterprise plan users. However, the US-based AI giant is still working to enable access to Fable 5 on AWS, Google Cloud, and Microsoft Foundry.

On the other hand, Anthropic has re-enabled Claude Mythos 5 access for select US organisations, after it received approval from the US government on June 26.