Gemini 4 Argon Goes Live: $2 Intro API Price, but Antigravity and AI Studio Are Ultra-Only

On October 13, 2026, Google made Gemini 4 Argon available to everyone, at least in name. The model was first shown a few weeks earlier, when Google said it would move with care because of the risk of misuse. Now it rolls out worldwide. Developers reach it through the API in AI Studio and the Gemini platform. Enterprises use Gemini for business. Consumers use the Gemini app and AI Mode in Search. Google says Argon is built for "deep work": multi-step tasks in coding, finance, legal work, cybersecurity and science. It also says the model ranks first on the Vals Index, which scores models on work from economic sectors such as finance, coding, legal and tax. The model has a 1M-token unified input-output limit and a new continuation protocol, so it can keep working on long tasks.

The price and access rules are the part to read closely. The API launches at an introductory $2 per million input tokens and $10 per million output tokens, with cached input at a 95% discount. That price ends on November 30. After that the rate is $4 and $20, which doubles the bill. Ars Technica tells developers to put this in their plans before they build workflows around the model. The subscription rules are tight too. In the Gemini app, Argon is only for AI Pro ($20 a month) and AI Ultra (from $99 a month). In Antigravity and AI Studio it is only for Ultra, because multi-threaded agentic coding uses a lot of compute. Google did not say when other tiers will get it, and it will keep Gemini 3.1 Pro for now. Free users, who have already lost most Flash access, get none of it.

Google published a long list of cases, and most of them are about agents that work for hours without help. It says that Argon in Antigravity raised agentic code submissions by more than 35% and cut rollbacks by nearly a third. At Chrome, Argon helped pull a roughly 6-million-line folder of browser features out of the 40-million-line Chromium codebase, across hundreds of files, in weeks instead of months. On the open-source Agent Development Kit, it worked alone overnight and cut in-memory session preparation by 40% in long sessions and sped up cold imports by 21%. On Gemma 4 E2B with vLLM, throughput went from about 37,900 to over 108,000 tokens per second, a 2.85x end-to-end speedup on an A100, and Google says the tokens are identical. Ars adds two more: removing redundant emoji rendering cut server load by 45% in the YouTube comment system, and Argon found bugs in the Project Wing drone simulator that had sat there since 2018.

The science and safety claims also come with numbers. Google says an arcsin routine written by Argon and formally verified in Lean cuts spacetime cost by more than 100x against a published quantum CORDIC baseline. It says Argon turned an internal AlphaQubit decoder into one optimized kernel and nearly doubled its throughput, and that a team including Professor Matt Landreman found a broader class of counterexamples to Grad's conjecture on stellarator fields. In private Gamow Labs evaluations, Argon scored 57% on RareBench and 52.2% on LabBench. On embodied reasoning it scores 87% on ERQA, and a humanoid robot sank seven cornhole shots in a row after Argon learned from video. For safety, external bio red teams unlocked under 1% of bio sub-tasks, violations fell by nearly half, shortcut attempts dropped to 2%, and attack success on the Gray Swan benchmark fell to 0.7%. Google will also give 10,000 scientists the top Antigravity tier free for six months.

Keep one fact in view: every number above comes from Google's own blog, and the Ars report adds no outside replication. The Vals Index rank, the private evaluations and the internal gains all need independent teams to test them on their own tasks. For most consumers the real change is that the best Gemini now costs money. For developers the question is the cost after the intro price ends, and whether Argon earns that price on long tasks. Try it on your own real work at small scale first. Then decide whether to change your workflow.