Anthropic Unveils Two New AI Models: Claude Fable 5.2 and Mythos 5.2

Anthropic has unveiled two new AI models: Claude Fable 5.2 and Mythos 5.2, in a move that pits them directly against OpenAI's GPT-6 Astra. This announcement comes just weeks after GPT-6 Astra was released.

The key selling point of these new models is their ability to offer higher performance at a lower cost than their competitors. Professionals who use AI models are increasingly weighing the price-performance ratio when deciding which model to choose, and Anthropic's new offerings aim to win them over with better value for money.

Anthropic's latest move is crucial as it tries to kick-start its revenue growth again after a slowdown in the third quarter was revealed in its recent SEC filing. The company needs this momentum to build before its planned IPO.

The pricing of Fable 5.2 and Mythos 5.2 reflects their competitive edge, with costs that are 20% lower than those of GPT-6 Astra for standard input and output tasks. For example, ordinary users will pay $8 per million input tokens and $40 per million output tokens for Anthropic's models, compared to $10 and $50 respectively for OpenAI's model.

Anthropic's pre-release testing of its new models has yielded impressive results, with the company reporting that these models have helped engineers complete complex software migrations, designed working electronic devices from scratch, and even made significant progress in robotics evaluations.

The Fable 5.2 and Mythos 5.2 models have reportedly outperformed their competitors on coding evaluation benchmarks, with Anthropic noting that over 90% of its own code is written by AI. This suggests a high level of proficiency in generating human-like text for programming tasks.

While OpenAI's Codex has gained ground in recent months due to lower pricing, Anthropic's Claude Code product remains a top choice among programmers. However, the company acknowledges that OpenAI's more affordable model may be attracting some users who prioritize cost over performance.

In areas requiring highly ambitious and open-ended technical work, Anthropic recommends using its latest Fable 5.2 and Mythos 5.2 models. These models are designed to tackle complex tasks where human expertise is often necessary, positioning them as competitors to OpenAI's Astra model in the high-end market.

To balance performance with cost, Anthropic suggests using its Opus 5.5 model for daily tasks. While more expensive than OpenAI's GPT-6.1 Sol, this model could still offer better value if it consistently produces high-quality results, leaving users to weigh their options and choose the best fit for their needs.

Anthropic's Claude Fable model has demonstrated impressive capabilities in solving complex mathematical problems, including a notable achievement in formalizing a proof within 30 hours using a team of 1,000 agents powered by Fable 5.2.

This is significantly faster than the 11 days it took for an unreleased research version of Claude to accomplish the same task, highlighting the efficiency and speed of Anthropic's model. The company is touting this achievement as a key benefit, allowing human mathematicians to focus on exploring novel ideas rather than spending excessive time checking proofs.

In addition to its mathematical prowess, Claude Fable has also shown promise in solving theoretical physics problems related to black holes, and identifying potential breakthroughs in gene therapies. These advancements demonstrate the model's versatility and ability to tackle a wide range of complex challenges.

Anthropic is also highlighting the results of an unnamed security company that found Claude Mythos 5.2 to be highly effective in identifying vulnerabilities in code, with one in six confirmed bugs being missed by other models tested. This suggests that Anthropic's model could have significant implications for cybersecurity and may offer a valuable tool for companies looking to bolster their defenses against potential threats.

Anthropic's latest models have demonstrated impressive capabilities in design and production, generating three functional electronic devices from a single prompt. The devices included circuit boards, 3D-printed enclosures, and software, showcasing the potential for artificial intelligence to drive innovation in hardware and robotics.

The company has also emphasized its focus on safety features within these new models. Specifically, they have implemented safeguards against high-risk cyber activities such as penetration testing, exploit generation, and binary-based vulnerability assessment. Additionally, research tasks in biology are largely blocked by the system.

Claude Fable 5.2 is now available on major cloud platforms including Amazon Web Services, Google Cloud, and Microsoft Azure. Meanwhile, access to Mythos 5.2 is restricted to application-only for cybersecurity professionals and life sciences researchers through Anthropic's platform.