Emad Mostaque · lab leaders · stability ☆ FOLLOW 31 topic(s) over 13 day(s) — inference, architecture, multimodal. on X ↗
CARD NO. 2026-10-04 Gemini 4 Argon supports 1M output tokens while Claude Fable 5.5 rolls out inference · 9 voices
CARD NO. 2026-10-03 Performance evaluations of Claude 5.5 and Fable 5.5 reveal reduced hallucination rates evaluation · 8 voices
CARD NO. 2026-09-30 DeepSeek Harness releases official TUI client alongside new agent benchmarks evaluation · 7 voices
CARD NO. 2026-09-28 Labs release new models and scientific agent benchmark leaderboards evaluation · 10 voicesGoogle launches new Gemini Flash audio model variants multimodal · 8 voices
CARD NO. 2026-09-26 Google introduces Gemini 3.8 Flash TTS and Flash-Lite TTS audio generation models multimodal · 16 voicesDeepSeek Training a 2T Model with Hardware Expansion Plans training · 6 voices
CARD NO. 2026-09-25 Anthropic releases the faster and cheaper Opus 5.5 model architecture · 11 voicesDeepSeek V4.1 Flash launched alongside plans for a 2T model training architecture · 9 voices
CARD NO. 2026-09-24 Google launches Gemini 3.8 Flash and Flash-Lite TTS audio model family multimodal · 10 voicesRelease of Ternary Bonsai 2 27B and the Qwen3.8 real-time translation model inference · 8 voicesDeepSeek advances large model development and outlines hardware training roadmap architecture · 6 voices
CARD NO. 2026-09-23 Google debuts Googlebook laptops alongside new multi-modal model releases multimodal · 10 voicesDeepSeek introduces V4.1-Flash and outlines future large model training plans inference · 8 voicesPrismML introduces Ternary Bonsai 2 27B quantized model inference · 6 voices
CARD NO. 2026-09-22 SpaceXAI launches Grok 4.7 and OpenAI expands Astra into legal tech evaluation · 18 voicesGoogle DeepMind launches Gemini 3.8 Live and audio reasoning models multimodal · 15 voicesNext-generation test versions of Claude and Grok spotted in development architecture · 12 voicesDeepSeek V4.1 Flash launched alongside 2T model training plans architecture · 7 voicesPrismML releases Ternary Bonsai 2 27B compressed from Qwen3.8 27B inference · 6 voices
CARD NO. 2026-09-21 Astra deployed for legal practice and engineering teams architecture · 14 voicesAnthropic has released a beta version of Salesforce integration into Claude, providing pre-built sales skills for managing accounts and opportunities. The company also reported that Claude now writes agents · 11 voicesDeepSeek releases the V4.1-Flash model for developers architecture · 6 voicesRelease of Ternary Bonsai 2 27B quantized model based on Qwen3.8 inference · 5 voices
CARD NO. 2026-09-20 GPT-6 Astra successfully decrypts historical radio transmissions never previously cracked evaluation · 9 voicesBolt platform integrates DeepSeek V4.1 Flash, GLM, and Kimi K3 into its model picker inference · 8 voicesQwen3.8 introduces quantized and real-time simultaneous translation models inference · 6 voices
CARD NO. 2026-09-19 GPT-6 Astra deciphered a 1918 German radio transmission and played Minecraft ai · 12 voicesPrismML launches Ternary Bonsai 2 27B based on Qwen3.8 and Qwen3.8-LiveTranslate ai · 7 voicesDeepSeek V4.1 Flash integrated into Bolt Forge and used for document processing ai · 4 voices