Models
Meta launched Muse Wonder and Muse Wonder Full on October 12. Developers can call Muse Wonder now. The fuller model is for Muse Code subscribers.
Why it matters A new frontier model is priced and available for agent work today. The Muse app does not have it yet.
Enterprise AI
Coinbase adapted the open-weight Qwen3.5-9B on its own Onramp fraud data. On the company's test it beat Claude Opus 4.5 and answered faster.
Why it matters A company can train a small open model on data only it has, and beat a frontier model on one job. Other companies can copy that path.
Models
OpenAI is rolling GPT-6 into ChatGPT, and a reply can include buttons, forms, charts, and small tools.
Why it matters A chat box that was a transcript can now be an interface the model writes, on free accounts as well as paid.
Models
Google DeepMind released a 740-million-parameter open model that embeds text, images, audio, and video on a device.
Why it matters Private photos, voice memos, and code can be searched without uploading them.
Agents & Automation
Odyssey released a world model you can move inside, with a public research preview.
Why it matters A learned world that can train a robot policy would matter. That claim is the lab’s, and this is not a robot you can deploy.
Research
Hugging Face released a gene-finding model and a database of 566 million predicted protein-coding loci.
Why it matters Most software teams can note it and move on. The number is a pile of candidates, not confirmed genes.
The same developments, with the facts and why they matter.
On October 12, Meta released Muse Wonder and Muse Wonder Full. Muse Wonder is on the Meta Model API and Muse Code at $4 per million input tokens and $20 per million output tokens. Muse Wonder Full is for paying Muse Code subscribers. The consumer Muse app comes later.
On October 8, Coinbase post-trained Qwen3.5-9B for Onramp fraud detection. The company says it beat Claude Opus 4.5 on every detection metric in its own benchmark, including a 9.6 percent higher F1 score, at 55 percent lower latency.
GPT-6 is rolling out in ChatGPT from October 7, and replies can include controls and small tools the model composes. Work and Codex stay on their previous models.
EmbeddingGemma 2 is an Apache 2.0 embedder small enough to run on a phone, across text, images, audio, and video. Sensitive retrieval can stay on the device that already holds the files.
Odyssey-3 is a public preview of a world model that generates a scene and predicts how it changes when someone acts. The robot and driving results are the lab’s, and they are not a system you can deploy from the post.
Hugging Face published Carbon-A and 566 million predicted gene loci across 22,617 species. Treat the loci as candidates with a confidence score. This does not change how a typical software team builds.