Your Guide to the AI Revolution

Ai2's AstaBrief 8B (October 2, 2026) is an Apache 2.0 model, fine-tuned from Qwen3-8B, that writes cited research reports. What it does and how to read it.

Cohere Embed 5 (September 30, 2026) ships Pro and Fast embedding models sharing one vector space, at $0.12 and $0.08 per million tokens. Specs and fit.

Sixteen notable releases from OpenAI, Anthropic, xAI, Cohere and the open-model labs, September 3 to October 7, 2026, with dates, licenses, pricing patterns and guides.

EmbeddingGemma 2 (October 6, 2026) is a 740M-parameter Apache 2.0 embedding model for text, code, images, audio and video. Specs, results and adoption.

Qwen-Image-2.1 (September 20, 2026) unifies image generation and editing with native transparency. Its weights are public, but the license is non-commercial.

Qwen3.8-Omni-Flash (September 18, 2026) takes text, image, audio and video input with a 1M context, API only. Capabilities, cost claims and limits.

Mistral Large 4 entered public preview on October 6, 2026: 1T parameters, 52B active, 1M context. Weights are promised for end of October. What's known.

H Company's Holo4 (September 28, 2026) brings open-weight computer-use agents in 27B and 35B-A3B sizes. Capabilities, results and the license split.

DeepSeek-V4.1-Flash (September 10, 2026) is a 552B MoE with open weights under MIT and a 1M context. Architecture, license, pricing and self-hosting.

xAI released grok-voice-transcribe-2.0 on September 17, 2026; version 1.0 now routes to it. API features, streaming turn detection and what to check.

xAI's Grok 4.7 (September 21, 2026) costs $2/$6 per million tokens under 200K-token prompts, with a 500K context. Pricing, availability and limits.

OpenAI's GPT-6 Sol and Luna (September 22, 2026) halved GPT-5.6 prices. Luna costs $0.10/$0.50 per million tokens. Pricing, caching and fit.