At 10:00 AM Beijing Time on November 10, 2026, Anthropic has released Claude Fable 5.2 and Claude Mythos 5.2. The company states that both products use the same underlying model but are configured with different security and bioscience domain protections, and employ different access conditions. Fable is now available to general users, while Mythos is only open to organizations that have passed cybersecurity or life sciences verification programs. Standard mode is priced at $8 per million input tokens and $40 per million output tokens. Officially, the company also offers a higher-priced optional fast mode.
Same underlying model yields two access conditions
Anthropic has configured Fable as a version that restricts more high-risk cybersecurity and bioscience tasks, and states that on certain intercepted topics, it will transparently forward processing to other Claude models. Mythos uses fewer restrictions and is aimed at reviewed dual-use work, thus the two cannot be written as products with identical access scopes. These two products are not separately trained base models; the official description differences focus on protection methods, available targets, and the open conditions of professional capabilities. Passing the verification program is a prerequisite for accessing Mythos, but this does not mean that the relevant capabilities have already obtained independent security certification.
In terms of capabilities, Anthropic claims that version 5.2 has improved coding, reasoning, visual, scientific, and engineering tasks, listing internal pre-release testing, partner trials, and several research cases. The company announcement also provides multiple benchmark and application results, but the current public information stream from leading media has not reproduced these experiments or formed independent capability rankings. Therefore, this article only treats the relevant conclusions as Anthropic's self-reported results, does not treat the company's tests as externally verified proof, and does not judge the actual performance of other models in similar tasks based on this.
Public Pricing and Company-Reported Speed Caps
The standard mode public pricing is $8 per million input tokens and $40 per million output tokens. The optional fast mode increases the corresponding prices to $16 and $80, with units still being per million input or output tokens rather than subscription monthly fees. Both product configurations use the same set of public unit prices. Anthropic did not provide a complete billing example for different tasks in this release material, therefore the public unit prices can be verified but cannot be used to infer the total cost or cost advantages for each individual project based solely on unit prices.
Anthropic claims that Fable 5.2 and Mythos 5.2 output speeds are up to 30% faster relative to version 5.1, with fast mode token generation speeds reaching up to 2.5 times that of normal mode. Both sets of figures are upper limits disclosed by the company and do not represent the same magnitude of improvement for every type of input, tool invocation, or workload. The official did not provide a set of speed comparison results replicated by independent institutions on the release page. Any related statements must retain company attribution and cannot be rewritten as fixed acceleration, stable latency, or guaranteed cost reduction.
Cloud Platform Channels and Official Product Tiering
Fable 5.2 is now available through the Claude developer platform and has entered Amazon Web Services, Google Cloud, and Microsoft Azure cloud services. Quick mode can be used in Claude code tools, the Claude developer platform, Claude model services on Google Cloud, and Microsoft Foundry environments. These are the launch channels listed by Anthropic, which do not necessarily represent identical quotas or contracts across all channels. Mythos and Mythos quick mode remain subject to verification program restrictions; Claude enterprise customers can also use Claude security services included in the model family according to official terms.
In terms of product tiering, Anthropic recommends that most coding and knowledge tasks begin with Opus 5.5, then select higher or lower tier models based on task scope. The company positions Fable and Mythos as options for handling open-ended technical work, and describes two orchestration approaches: acting as a consultant to seek advice from other Claude models, or serving as a coordinator to delegate simpler tasks to other models. These constitute officially recommended usage patterns, not uniform processes adopted by all customers, nor can they be used to infer that multi-model orchestration is faster or cheaper for every task.
Security Assessment Boundaries and Undisclosed Information
Anthropic stated that pre-launch security work included alignment training and testing, abuse prevention, behavioral monitoring, and external evaluations. The evaluating institutions involved included the UK AI Safety Institute, Frontier Design, and METR. The company simultaneously clarified that these external institutions had not reviewed published blogs or system card content, nor did they necessarily agree with the assertions contained therein. Participation of external institutions in testing cannot be rewritten as endorsement of all performance, security conclusions, or product policies. Official judgments regarding high-risk behaviors still fall under the company's own assessment.
As of this round of verification, official materials have not disclosed model parameter scale, training compute, context window, complete regional restrictions, customer count, or revenue impact, nor have they provided benchmark retesting that could be independently verified by this round's media metadata. The Verge and TechCrunch's authorized public information streams respectively recorded product launches, timing, and improvement directions proposed by the company, but the main text did not use their full article content. What can be confirmed is the launch, pricing, channels, and access structure; capabilities, speed, and security performance remain handled according to Anthropic's self-reported data.