#model-releases

DeepSeek V4.1 Flash adds vision and lower prices. Compare 19 official benchmarks, agent frameworks and API costs; V4 Pro now continues past September 14.

A dated ledger of AI model releases in September 2026, each row verified against the vendor’s announcement, with price, context and what it replaces.

In the 2025-26 record we checked, two labs gave a public reason before holding back open weights. What a stated delay tells a buyer, and what it does not.

xAI's Grok Bot launched in beta on August 11. The launch page says Bots have their own computer; xAI's own docs say an account's Bots all share one.

Meta released Muse Glimmer 30B under a genuine Apache 2.0 license: a dense, multimodal model distilled from Muse Spark and aimed at local agent work.

NVIDIA's 30B, 3B-active Nemotron 3.5 Lightning trails Qwen 3.6 35B A3B on 13 of 14 rows in NVIDIA's own table. The pitch is speed, not peak accuracy.

Qwen3.8-Max ships as a 2.4T-parameter MoE with 95B active and a 1M context. What the vendor benchmark table shows, and what was only promised at launch.

DeepSeek V4 Flash exits preview into public beta as the 0731 checkpoint: same 284B architecture, vendor-stated agent benchmarks, no weights posted yet.

MiniMax launched H3, a 2K video model with native synced audio at $0.13 per second. Weights are promised within days, not published yet.

ByteDance launched Seedance 2.5 on Jimeng and Doubao Pro with 30-second one-take video. Its own post says the BytePlus ModelArk API is still coming.

Twelve AI models launched in a single week of March 2026 from OpenAI, Google, Mistral, xAI, and more. Developer guide to capabilities, pricing, and selection.