On August 31, 2026 DeepSeek published V4-Flash-Vision-Exp, a roughly 305B-parameter experimental multimodal model, on Hugging Face under the permissive MIT license. It shows substantially improved agent benchmarks over prior V4-Flash versions.
DeepSeek ships experimental V4-Flash-Vision multimodal model on Hugging Face
Teams building vision or agentic features can experiment with an MIT-licensed multimodal model at no cost, but should treat an 'experimental' release as a prototype rather than production-ready.
Source: AI Weekly
More that helps you.
Tencent open-sources Hy4, a 770B model with 1M context under Apache 2.0
On August 28, 2026 Tencent released and open-sourced its Hunyuan Hy4 preview: a 770B-parameter Mixture-of-Experts model activating 49B parameters per token, with a 1-million-token…
Apple opens English public beta of rebuilt, Gemini-powered Siri
On September 15, 2026 Apple opened the English public beta of its rebuilt Siri, powered by custom models built with Google Gemini and split between on-device and Apple's Private Cl…
Anthropic ships Claude Fable 5.1 and Mythos 5.1, cutting agent costs ~45%
On September 2, 2026 Anthropic released Claude Fable 5.1 (broadly available) and Mythos 5.1 (restricted to vetted organizations), citing about 45% lower costs for agentic workloads…
Z.ai ships GLM-5.3-Flash: 1M-context multimodal coding model at ~1/10 the price
On Aug 26, 2026 China's Z.ai released GLM-5.3-Flash, a 320B-parameter (18B active) mixture-of-experts model with a 1M-token context and native image/video input, under an MIT licen…
Get briefs like this tuned to you.
In the app, Founder Briefs are personalized to your country, industry and stage, and you can save the ones that matter.
See plans →