Models
Canva launches Craft, its first in-house image model, which plans a layout before it generates
Trained on human-made design templates, Craft aims for readable text and editable posters, slides and social graphics.
Models
Model releases, capabilities, and platform changes from the frontier labs and open ecosystems.
- Models
Anthropic launches Claude Duet, an AI agent for professionals and small business owners
The agent is available as a research preview and competes with Meta's Muse and OpenAI's Dots.
- Models
Recursive releases Flint, a free laptop coding model it says was built largely by its automated research system
The startup says its Eureka Machine helped with nearly every development step, so a small team could build the model in under four months.
- Models
Google open-sources TorchTPU to run PyTorch apps on its own AI chips, a day after Amazon's similar move
Developers can move existing PyTorch apps to Google's TPUs with a configuration change instead of a rewrite.
- Models
MiniMax releases M3.1 Flash and says it cuts the cost of agent tasks by 70 percent
The 428-billion-parameter update to M3 runs nearly three times faster, and MiniMax says it was the anonymous model known as Space Bunny.
- Models
OpenAI's 'Aurora' models resurface on Arena generating images, pointing to a new image model
After the three model names were briefly listed in OpenAI's API, they turned up in blind arena tests producing pictures.
- Models
Three unannounced 'Aurora' model names reportedly appear in OpenAI's API, but there is no sign they work
A post on X listed aurora-blue-alpha, aurora-green-alpha and aurora-red-alpha without showing any successful request.
- Models
DeepSeek releases open-weight V4.1-Flash-1017 and opens its search engine to developers
The company says the update beats Claude Opus 5 and GPT 5.6-Sol on its internal coding and agent benchmarks while producing shorter outputs.
- Models
Meta releases the weights for Muse Spark Open, an 816-billion-parameter model built to run on two Mac Studios
Meta calls it the most capable US open-weight model to date. Running it locally takes at least 1 terabyte of RAM.
- Models
OpenAI reportedly plans one or two new models before year end, but its next flagship still needs alignment work
According to The Information, the flagship uses the same looped transformer technique as Astra, which makes safety monitoring harder.
- Models
Google releases Gemini 4 Argon worldwide as its new frontier model, priced at $2 per million input tokens
Argon replaces the Gemini Pro line and ranks first on the Vals Index. Free and Plus users stay on Flash models, and Google is widening its Fairwind cyber-defender program.
- Models
Meta releases Muse Wonder and Muse Wonder Full, its first attempt at frontier models
Meta says the pair score about as well as Fable 5.1 and GPT-6 Astra on its own benchmark tests. Wonder costs $4 per million input tokens through Meta's API, while Wonder Full is only available to Muse Code subscribers.