Nathan Lambert · researchers ☆ FOLLOW 25 topic(s) over 10 day(s) — architecture, inference, evaluation. on X ↗
CARD NO. 2026-10-05 Google launches Gemini 4 Argon with a 1 million token output limit architecture · 23 voicesOpenRouter adds new models and reports high usage share for Claude Opus 5.5 inference · 5 voicesGLM models expand availability across Cursor and Hugging Face ecosystems inference · 4 voicesOpenAI reports model extraction campaign linked to Moonshot AI developers evaluation · 4 voices
CARD NO. 2026-10-04 Google launches Gemini 4 Argon model for complex software engineering tasks architecture · 24 voicesFireworks AI introduces Ember-1 built on Kimi K3 with reduced token usage inference · 10 voicesOpenRouter integrates new decision models and highlights platform usage trends inference · 5 voicesZhipu AI releases GLM 5.3 model family including Flash and Max variants architecture · 4 voices
CARD NO. 2026-10-03 Codex adds premium speed tiers and cloud environments system_design · 15 voicesGoogle releases Gemini 4 Argon with a 1M token output limit architecture · 14 voicesOpenRouter adds Pareto 26.10 Preview model inference · 4 voices
CARD NO. 2026-10-02 Multiple new AI models integrate into the OpenRouter platform inference · 4 voicesAnthropic publishes research on GLM model exploit capabilities evaluation · 3 voices
CARD NO. 2026-10-01 OpenAI releases GPT-6.1 and patches vision bugs across GPT-6 models architecture · 21 voicesGoogle launches Gemini 4 Argon and Gemini 3.8 Flash TTS audio models architecture · 14 voicesOpenRouter integrates specialized decision models for coding agents agents · 5 voicesInquiry into reasoning extraction campaign and orbital TPU test mission system_design · 4 voices
CARD NO. 2026-09-30 Grok 4.7 claims top ranking on the AA cyber index evaluation · 22 voicesDebates over security risks associated with open-source models evaluation · 3 voices
CARD NO. 2026-09-29 Opus 5.5 released with distillation and reinforcement learning from a teacher model training · 4 voicesResearchers debate security risks associated with open-source AI models evaluation · 3 voices
CARD NO. 2026-09-28 Introduction of the training and distillation approach for Opus 5.5 training · 4 voicesDebate intensifies over security risks and the role of open-source AI models evaluation · 3 voices
CARD NO. 2026-09-27 New research explores distillation and reinforcement learning for agents training · 4 voices
CARD NO. 2026-09-25 SpaceAI releases Grok 4.7 with advanced coding capabilities and high performance architecture · 5 voices