New Models
Claude Fable 5.2
Anthropic launches Claude Fable 5.2 and Claude Mythos 5.2
Claude Fable 5.2 and Claude Mythos 5.2 are the same underlying model: Fable ships on paid plans and the API, Mythos stays restricted to approved programs. Fable 5.2 costs $8 / $40 per million tokens, down 20% from $10 / $50, Anthropic says it is about 30% faster, and it is #1 on the Artificial Analysis Intelligence Index at 65. The system card flags stronger covert capability and more evaluation awareness.
$8 / $40 per 1M input / output tokens65 Artificial Analysis Intelligence Index~30% faster
New Models
Clairic
Cloudflare's Clairic extraction model, Apache 2.0 coming
Cloudflare's Clairic extraction model. Its Apache 2.0 release is still coming, so it is not open source yet. Michelle Chen of Cloudflare's AI Platform explained on the show why Cloudflare now calls itself a neolab.
New Models
Gemini 4 Carbon (GA)
Gemini 4 Carbon is generally available, #1 on CursorBench
Carbon, Google's top model above Argon, went GA. The intro price is $2 / $10 per million tokens until November 30, then $4 / $20. At high effort it is #1 on CursorBench at 60.4%, just ahead of Fable 5.2 at 60, and it can write up to a million output tokens in a single pass.
60.4% CursorBench (high effort)$2 / $10 per 1M tokens until Nov 301M output tokens in one pass
New Models
Gemini 4 Hydrogen
Google launches Gemini 4 Hydrogen, live on ThursdAI
A new fast Gemini 4 model, announced live on the show by Ivan Leo of Google DeepMind. Ivan says it is competitive with the original Gemini 4 Argon on the Artificial Analysis index and scores higher at maximum thinking, runs about 2.5x faster than Gemini 3.8 Flash at roughly the same price, and calls multiple tools at once. Google's own numbers.
~2.5x faster than Gemini 3.8 Flash
New Models
Muse Ember
Meta launches Muse Ember, its smallest Muse model
Meta launched Muse Ember, its smallest Muse model, with closed weights.
New ModelsOpen weights
MiniMax H3-Image
MiniMax releases H3-Image with open weights
The image version of MiniMax's H3 video model: 4K images (16 megapixels) in about 10 seconds, up to 16 reference images, editing, and open weights on Hugging Face. The license still needs a look.
4K 16-megapixel images~10 s per image16 reference images
New ModelsOpen weights
Arete-27B
Nous Research releases Arete-27B
Qwen3.8-27B post-trained with supervised fine-tuning toward what Nous calls Western values. On Nous's own 803-prompt eval the propaganda rate went from 22% to 0.9%, with 65.6 on Hermes Bench against 65.8 for the base model. Apache 2.0, with the eval and the data released too.
22% → 0.9% propaganda rate, own eval65.6 Hermes Bench (base 65.8)803 eval prompts