RX-7 model testModel Guides
GPT-6.1 Terra, Sol and Claude Opus 5.5 build the same RX-7 browser demo
MattVidPro compares RX-7 models, rotary engines and driving controls, then weighs the results against model prices.
Find your next discovery
Model Guides
New models, clear comparisons and choosing a model for the task.
RX-7 model test
AI drug researchModel Guides
Anthropic acquires Mithril Medicine to expand AI-assisted drug research
The acquisition brings early muscular-dystrophy research into Anthropic, with mouse results and a conditional late-2028 clinical-trial goal.
IPO cash positionModel Guides
Anthropic’s S-1 puts its cash position at the centre of the IPO debate
The filing reports $112.4 billion in cash and equivalents at June 30, while the proposed offering remains unfinished.
Choose the next stepModel Guides
CLM-8B: an open model that scores possible agent actions
Last Week in AI explains an alternative to generating a long answer for every small decision.
Learning to improveScience & Research
Recursive self-improvement: five levels of autonomy and the Dream-RSI approach
Last Week in AI explains two papers about how AI systems could improve the way they learn and explore.
Chat gets a faceVoice & Music
Gemini 3.8 Live Avatar gives the AI conversation a visible face
Last Week in AI discusses Google’s real-time avatars for enterprise conversations.
Open weightsModel Guides
Meta releases Muse Spark Open weights for coding and agent workflows
Muse Spark Open combines multimodal input with downloadable weights and demanding self-hosting requirements.
Connected voice toolsVoice & Music
Voxtral 3 connects transcription, speech and voice agents
Mistral combines speech models with a developer SDK for agents that can listen, respond and use software tools.
More for lessModel Guides
GPT-6.1 Sol and Claude Sonnet 5.5: more capable everyday AI at lower cost
Last Week in AI discusses two model launches focused on faster work and lower costs.
Check the claimScience & Research
TasteVal and AI mathematics: measuring research progress without overstating it
What mathematical manuscripts and TasteVal actually establish—and why benchmark gains are not complete research automation.
AI on deviceModel Guides
Liquid AI LFM3: a multimodal model family for on-device use
Liquid AI has introduced the compact LFM3 family for text, vision and audio tasks.
Learn. Keep.Science & Research
Google SLEEP: research into retaining knowledge in AI models
Google Research describes SLEEP, a method intended to help models learn new information while retaining earlier knowledge.