™All Models & Products
-
Illumina Launches SpliceAI2 Genomics AI Model, Lifting Disease-Variant Detection by 17%
Why it mattersOn rare-disease datasets SpliceAI2 detected 17% more disease-related variants than comparable models, and it was trained on more than 100 times the data of the 2019 original; code and weights are open for academic, non-commercial use.
-
Google Opens SynthID Detector Worldwide: Check Any Photo for AI Watermarks
Why it mattersThe Detector was previously limited to a few journalists and researchers; now anyone who signs in can upload files to check.
-
OpenAI Rolls GPT-6 "Intelligent UI" to All ChatGPT Tiers: Answers Become Interactive Interfaces
Why it mattersChatGPT's main chat box, used by 1.2 billion people weekly, moves from text answers to interactive interfaces for the first time; GPT-6 Sol and Luna already reached the API and Codex on September 22, and Google shipped a similar feature for Gemini in May.
-
Claude vs GPT for Coding: Same Mid-Tier Price, a 2.5× Flagship Gap, Benchmarks Measured Differently
Why it mattersMid-tier Sonnet 5.5 and GPT-6 Sol are priced identically ($2 input / $10 output per million tokens); flagship Opus 5.5 is about 2.5× cheaper than GPT-6 Astra, while entry-tier GPT-6 Luna is ten times cheaper than Haiku 4.5.
-
OpenAI Publishes 722 AI-Generated Math Manuscripts Citing Riemann Hypothesis Progress; None Peer-Reviewed
Why it mattersIt is the largest of OpenAI's math releases: 722 manuscripts across 372 problem families, none peer-reviewed; the internal model behind them is unreleased, and no prompts or compute costs were published.
-
Google Launches EmbeddingGemma 2: 740M-Param Multimodal Embedding Model Goes Apache 2.0, Runs On-Device
Why it mattersThe first generation passed 20 million downloads in a year; the new one puts text, code, images, audio and video in one vector space under Apache 2.0, with the text-only build at about 191MB of RAM.
-
Mistral Debuts “Le Chonk”: Open-Weight Large 4 in Public Preview, Weights Drop Oct 27
Why it mattersIt was trained on about 4,000 Nvidia Grace Blackwell GPUs in Mistral's own European data centers; weights land October 27, and until then no third party can independently verify its performance claims.
-
Reka Launches Rho-1, a 19B Omni-Model Handling Text, Images, Video and Robot Actions in One Network
Why it mattersRho-1 understands and generates text, images and video and outputs robot actions inside one 19B-parameter network, with no tool calls; it is only a research preview, with no weights, API or third-party evaluation.
-
Nvidia-Backed Reflection Debuts Beam, Its First Open-Weight Frontier Model: 501B Params, 23B Active
Why it mattersBeam has 501B total parameters with 23B active, and the company says it needs a third to a quarter of GLM-5.2's inference compute; scores are self-reported and weights open later in October.
-
Google Trims Free Gemini: Only Flash-Lite Left From October 9
Why it mattersFrom October 9 free users get only Flash-Lite and AI Plus loses Pro, while AI Pro gains Deep Think, previously Ultra-only; Gemini reportedly has over a billion monthly users.
