Gemini 4
Gemini 4 Argon is live. Google made it generally available on the Gemini API on Oct 13 as gemini-4-argon, and our sweep fired at 15:45 UTC. It costs $2 in / $10 out per million tokens through Nov 30, then $4 / $20 from Dec 1, with cached input at $0.10 (then $0.20). The 1M-token context window is shared by input and output: replies default to 65K tokens and run up to 1M with long decode. It is not on the Free tier or the Google AI Pro plan. Google’s own table puts it ahead of Claude Opus 5.5 on 14 of 18 shared benchmarks; Artificial Analysis scores it 52.6, behind Opus 5.5’s 57.6. Gemma 4 is still a different product: Google’s open-weight line.
Leaks & news, newest first.
Gemini 4 Argon goes GA on the Gemini API
Google’s changelog lists gemini-4-argon as generally available, “our most capable model built for complex coding, agentic, math, and science tasks.” $2 / $10 per million through Nov 30, then $4 / $20; 1M context; knowledge cutoff March 2026 per the model card. Our sweep fired at 15:45 UTC. Our launch story. via ai.google.dev · deepmind.google
Argon signs a public GitHub commit through Google’s Antigravity agent
A commit in the open-source BlocSignal repository ends with “Co-authored-by: Gemini 4 Argon” at Antigravity’s agent address; earlier commits there name Gemini 3.7 and 3.8 Flash the same way. Likely a trusted-tester account; Argon is still not on the Gemini API. via GitHub · spotted by @Mr_Salio
Google announces Gemini 4 Argon, trusted testers first
Koray Kavukcuoglu’s launch post: rolling out to cyber defenders and trusted testers through the Fairwind Program, then paid API customers and Google AI Ultra. Introductory price $2 / $10 per million, cached input 95% off, 1M output tokens. Google’s own benchmark claims include 77.9% on DeepSWE v1.1; Vals lists it #1 on its index at 68.90%. via blog.google
DeepMind: Gemini 4 is in post-training, release “as soon as possible”
DeepMind chief Koray Kavukcuoglu told The Information’s AI Agenda summit that Google intends to release an early post-training version soon, and hopes to ship much earlier than the end of 2026. No date was given. via 9to5Google
Google: the Gemini 4 pre-training run has started
In the post launching Gemini 3.6 Flash, Google DeepMind wrote: “We have started our most ambitious pre-training run yet, for Gemini 4, and are excited by the progress.” The same post said 3.5 Pro was “currently testing with partners.” No date for either. via Google
Gemini 3.5 Pro slips to a staged July rollout — a “ground-up rebuild”
Reports put 3.5 Pro in a whitelist preview around Jul 8–12 with wider access ~Jul 17, citing a full re-architecture, a reported 2M-token context window and a “Deep Think” reasoning layer. The same reports mention a “Gemini 4 Flash” variant in development — anonymous sourcing, no Google confirmation. via TechTimes · BigGo (unconfirmed)
DeepMind departures raise pace questions
Fortune reported notable exits — including John Jumper leaving for Anthropic — prompting doubts about DeepMind’s pace in the model race. Context for why the 3.x line is stretching. via Fortune
Google I/O ships Gemini 3.5 Flash — and no Gemini 4
The Pichai-led keynote launched Gemini 3.5 Flash in preview and promised 3.5 Pro “next month.” Every “Gemini 4 at I/O” piece beforehand was speculation — the model shown was 3.5. via Fast Company · Google
Gemma 4 ships — and it is not Gemini 4
Google’s open-weight Gemma line hit version 4 for reasoning/agentic workloads — routinely mis-cited as “Gemini 4” in search results. Different product line entirely. via Google
The official word.
"Our most capable model built for complex coding, agentic, math, and science tasks."
Gemini API changelog, on gemini-4-argon going GA · Oct 13, 2026 — announced Sep 30 for trusted testers through the Fairwind Program; the pre-training run started by Jul 21.
$2 / $10 per million tokens, on Google's own API.
How much does Gemini 4 cost?
Google's own API lists Gemini 4 Argon (gemini-4-argon) at $2 per million input tokens and $10 per million output tokens, and $0.10 per million cached input tokens. Prices as published in the models.dev catalog, as of 2026-10-27.
What is Gemini 4's context window?
1.05M tokens, with up to 33K output tokens per response, as Google lists Gemini 4 Argon (gemini-4-argon).
Listed by 26 providers, exact ids below.
26 providers list Gemini 4 in the models.dev catalog (as of 2026-10-27), under the model ids shown below.
gemini-4-argongemini-4-argongoogle/gemini-4-argongoogle/gemini-4-argongemini-4-argongemini-4-argongemini/gemini-4-argongoogle/gemini-4-argongoogle/gemini-4-argongemini-4-argongemini/gemini-4-argongemini-4-argonAlso listed by Eden AI, GitHub Copilot, Kilo Gateway, LLM Gateway, Merge Gateway, NanoGPT, Neon, Ofox, OpenCode, OpenCode Go, Perplexity Agent, Prime Inference, Requesty, Venice AI.
Listings cover the model family, so a provider may list a sibling tier under the same name. Check the id before you build on it.
Head to head, on the public leaderboards:
Hear it the second the next Gemini answers.
We sweep Google’s public API every 60 seconds. Gemini 4 Argon fired at 15:45 UTC on Oct 13; the hunts for the next Gemini 4 and for Gemini 5 are armed. The moment one returns a real answer, a text or a phone call fires on paid tiers; free email + Chrome push follow about 15 minutes later. Not the rumor mill: the alert means it’s answering.
By joining you agree to the ArtificialWatch Terms of Service and Privacy Policy. You'll get the alerts you picked; unsubscribe in one click. 18+.