Search
Have an existing account?
Sign In
Tag: Model releases
Baidu’s new OCR model reads books without forgetting the start
Unlimited OCR 2 uses sparse attention to hold decoding cost flat across documents hundreds of pages long.
Canva trained its own image model on a decade of templates
Craft plans a layout before it renders a pixel, a two-stage design that Canva says puts it second…
Moonshot ships a coding model with weights and a speed tier
Kimi K3.1 Pro is a post-trained K3 with a million-token window and a faster sibling that Moonshot says…
Fireworks retrains Kimi K3 to stop burning tokens on itself
A derivative model cuts reasoning overhead by roughly 40 percent while holding accuracy steady, and it ships only…
Anthropic pushes Sonnet 5.5 as a cheaper everyday workhorse
Anthropic's mid-tier Claude gets a speed and cost overhaul, and it arrives with the same cyber safeguards the…
Grok 4.7 lands cheap but trails the frontier on agentic coding
SpaceXAI's newest model undercuts Western rivals on price while landing mid-pack on independent benchmarks.
StepFun’s Step 5 Preview stacks 92 layers to stretch agent runs
The Chinese lab squeezed 600B parameters into a sparse model that activates 27B per token and holds context…