AI News Roundup: GPT-6 Goes Full, Gemini 4 Argon Ships, and the Open-Weight Flood
This week the frontier labs stopped teasing and started shipping. OpenAI put GPT-6 on every ChatGPT plan — paid users got GPT-6 Sol, free and Go users got GPT-6 Luna — and turned on an Intelligent UI that streams charts, forms, and little calculators inside the chat instead of just text. The 500k context on GPT-6.1 Sol is the quiet part that matters most for anyone doing real work.
Google's Gemini 4 Argon reached general availability on October 13 after a early window limited to vetted cyber defenders. One million output tokens, a $2/$10 per million intro price that climbs after November, and it's now inside the Gemini app for Pro and Ultra subscribers. Anthropic answered with Claude Haiku 5.5 on October 7, roughly 75% cheaper than the model it replaces and built for the high-volume, low-stakes calls that eat most budgets.
The bigger story is the open-weight wall. Meta opened Muse Spark Open on October 15, an 816B mixture-of-experts model you can actually download and run, and Mistral's Large 4 — a trillion-parameter multimodal model — is set to drop its weights later this month. StepFun, Liquid, and NVIDIA all joined the 'run it yourself' pile this week. For teams that can't send data to someone else's API, that changes the math.
Creatively, Krea 3 went beta with reference-driven editing, Kling 4.0 pushed native clips to 30 seconds, and Ideogram 4.5 tightened local image edits. Suno started bundling spoken narration with original music in one track. The tools aren't just getting better; they're getting specific.
The counterweight was legal. Sony Music asked platforms to pull more than 260,000 tracks that imitate its artists, and Disney sent cease-and-desist letters over AI-generated video that borrowed its characters. The 'move fast' phase of generative media is colliding with the 'who owns this' phase, and this week you could see both at once.
Updated October 28, 2026: pricing and availability below were verified at publication.
Frequently asked questions:
Q: Which GPT-6 model do I get?
A: Paid users get GPT-6 Sol; free and Go users get GPT-6 Luna. The 500k context on GPT-6.1 Sol is the quiet part that matters most for real work.
Q: What did Gemini 4 Argon cost?
A: $2 in / $10 out per million tokens as an intro price that climbs after November, with a one-million-token output ceiling - now inside the Gemini app for Pro and Ultra subscribers.
Q: Which big models got open weights?
A: Meta opened Muse Spark Open on October 15, an 816B mixture-of-experts model you can download and run - and Mistral's trillion-parameter Large 4 promised weights by month's end.