The Memo - Special edition - Claude Fable 5.2 and Claude Mythos 5.2 - 9/Nov/2026
To: US Govt, major govts, Microsoft, Apple, NVIDIA, Alphabet, Amazon, Meta, Tesla, Citi, Tencent, IBM, & 10,000+ more recipients… From: Dr Alan D. Thompson Sent: 9/Nov/2026 Subject: The Memo - AI that matters, as it happens, in plain English AGI: 99% ASI: 2/50Anthropic releases…
Annons
Annons
To: US Govt, major govts, Microsoft, Apple, NVIDIA, Alphabet, Amazon, Meta, Tesla, Citi, Tencent, IBM, & 10,000+ more recipients… From: Dr Alan D. Thompson Sent: 9/Nov/2026 Subject: The Memo - AI that matters, as it happens, in plain English AGI: 99% ASI: 2/50Anthropic releases Claude Fable 5.2 and Claude Mythos 5.2Once again, we have this out to The Memo readers within a few hours of model release.Just seven months ago, Anthropic told us that Mythos Preview was too dangerous to release. That model found bugs in every major operating system and every major browser, and it was locked away (and only allowed out with a select group of trusted partners). Today, the same class of model in Fable 5.2 is available to the public, at less than a third of what Mythos Preview cost the Glasswing partners.Fable 5.2 and Mythos 5.2 are the same model with two different sets of safeguards. Fable 5.2 has strong cybersecurity and biology safeguards, and is available to everyone. Mythos 5.2 has some of those safeguards removed, and is only available to vetted organizations through Anthropic’s Cyber and Life Sciences Verification Programs.As with earlier models including Mythos 5, Fable 5, and GPT-6 Astra, both new models should be considered proto-ASI. You can view my 50 lagging ASI indicators at LifeArchitect.ai/ASI.Here are the points that matter most, in plain English.Claude now writes most of Claude. Anthropic says that over 90% of the changes merged into its core codebase are now written by Claude, with human guidance. Fable 5.2 solved two open problems in black hole physics. The first completes a calculation started in 2003, and finds that the transition to black hole formation is also abrupt in very small universes. The second finds the exact limit of a technique physicists have used for more than 15 years to rule out candidate theories of quantum gravity. Hawking-Page paper, modular bootstrap paper.Fable 5.2 designed three working electronic devices from a one-paragraph prompt. Working with Diode, a company that designs and manufactures circuit boards, the model designed a home voice-assistant hub, a USB speakerphone, and a handheld virtual pet. It did the circuit boards, the 3D-printed enclosures, and the software. All three devices worked without any electrical changes, and each board was ready for manufacturing in under a day. https://www.anthropic.com/research/claude-electronics-design and watch the video (link):Fable 5.2 controls physical robots, even on tasks it was never trained for. Anthropic says Fable 5.2’s ability to control physical robots is much better than earlier Claude models. That includes robotics tasks it was never explicitly trained on. In Anthropic’s demo, robots controlled by Fable 5.2 solve a puzzle cube, unscrew a bottle cap, and draw a picture. On the RoboDojo Sim benchmark, Fable 5.2 scores 64.9%, more than double Fable 5.1 (27.6%) and GPT-6 Astra (29.8%), and well ahead of Opus 5.5 (39.4%).Fable 5.2 wrote a new operating system. As an experiment, an Anthropic engineer asked Fable 5.2 to design a prototype operating system, starting from an empty folder. The model coordinated more than 1,000 subagents working around the clock for eight days. The result boots on a laptop, and includes a kernel, disk, keyboard, and network drivers, file storage, windows, a login screen, an installer, and built-in Claude features.1,000 Fable 5.2 agents formalized Fermat’s Last Theorem in 30 hours. An unreleased research version of the model took 11 days. Opus 5.5 with the same setup finished only 26% of the formalization in 18 hours. https://www.anthropic.com/research/formalizing-fermats-last-theoremMythos 5.2 modeled 28 of 34 whole virus capsids. The model accurately modeled 28 of the 34 adeno-associated virus (AAV) capsid structures it attempted, using two open-source protein-folding models. AAV capsids are the shells used to deliver gene therapies. Watch the video (link):Anthropic held back the capabilities that speed up AI itself. During training, Anthropic took steps to limit the research capabilities that most directly accelerate AI development. In accelerator kernel optimization and LLM architecture improvement, Fable 5.2 and Mythos 5.2 perform at about the level of Opus 5.5. Anthropic says this did not affect agentic coding, hardware development, or traditional machine learning research. This is the first time I have seen a frontier lab publicly say it deliberately slowed one of its own models in the area that matters most for recursive self-improvement.Performancehttps://lifearchitect.ai/models-table/?tab=rankingsClaude Fable 5.2 scores HLE=74.2% (with tools), well above Opus 5.5 (67.7%) and GPT-6 Astra (57.2%). In Jun/2026, Anthropic (Mythos paper, p258) noted ‘We consider GPQA Diamond to be a saturated evaluation and plan to stop reporting the performance of future models on it.’ Based on the highest testing suites we have, and the estimated ceiling for HLE due to errors (HLE≈51.3%), this model continues to meet the HLE half of my criteria for an early ASI system. (‘It is likely that any model with a primary score at >50% on HLE, and a secondary score at >90% on GPQA is an ASI system.’ https://lifearchitect.ai/asi/)Size estimatesGeneral model size is no longer an indicator of performance, but I still find it interesting. With all model details kept confidential, plus added complexity in reasoning/thinking mode, it is more challenging than ever to estimate token and parameter counts.Fable 5.2 is a point release on the same Mythos-class base as Fable 5.1, so my estimates are carried forward from that model.Pricing as a sizing signal. Fable 5.2 is priced at $8 per million input tokens and $40 per million output tokens, down from $10/$50 for Fable 5. An optional fast mode generates tokens up to 2.5x faster for $16/$80. Outputs are also up to 30% faster than Fable 5.1 and Mythos 5.1 in standard mode. A price cut alongside a speed increase on the same base points to inference efficiency gains rather than a smaller model.Alan’s estimates for Claude Fable 5.2 and Mythos 5.2:Total parameters: ~10T (MoE), the same Mythos-class base as Fable 5.1.Active parameters per token: centrepoint ~150B.Training tokens seen: centrepoint ~250T, with even more post-training reinforcement learning compared to Fable 5.1.Tokens per parameter: ~25:1.Dataset: A proprietary mix of public web data, public and private datasets, and synthetic data generated by other models, with extensive RL on long-horizon agentic tasks. Full subscribers can read how frontier labs are using extended datasets as well as synthetic data to train today’s models in my GPT-5 paper.Behind the curtain: Models 3 and 4Fable and Mythos are no longer frontier models.Today, Anthropic said it held back Fable 5.2’s ability to speed up AI research, keeping it at about Opus 5.5 level. Of course, Anthropic’s internal models don’t have that limit, and they have leapt well past anything publicly available.METR’s latest assessment of AI R&D acceleration at Anthropic (12/Aug/2026 to 29/Oct/2026) covers two unreleased internal models, called Model 3 (finished training Aug–Sep/2026) and Model 4 (finished training Sep–Oct/2026). For the most capable internal model in Oct/2026, METR reports average self-reported uplift of about 12x. In plain English, Anthropic’s own staff say they’re getting about a year’s worth of work done every month.In The Memo Special edition 7/Apr/2026 (Mythos), I noted that one of the major leading indicators of ASI is that the most capable models would no longer be publicly available. Models 3 and 4 are a clear example of this indicator.Full subscribers can read my estimates for Model 3 and Model 4 below, and see how they compare with Fable 5.2… Read more
