Google made Gemini 4 Argon broadly available on October 13, 2026, after first announcing it on September 30 for trusted cyber defenders through its Fairwind program. The release expands access to developers, enterprises and eligible subscribers—but availability depends on the product and plan, and benchmark results are not a universal verdict on which model is best.
3
20
43
49
Where Gemini 4 Argon is available
Google announced worldwide access for developers and enterprises through the Gemini API and Gemini platform in AI Studio. Google AI Ultra subscribers can use Argon in Antigravity; Google AI Pro and Ultra subscribers can use it in the Gemini app and AI Mode in Search.
43
For developers, the stable API model ID is gemini-4-argon. Google’s developer announcement also lists Android Studio as a way to build with Argon through the paid Gemini API tier.
17
33
The API and consumer subscription routes are distinct. Google says Argon is not available on the API Free tier or the Google AI Pro plan as API access; its launch announcement separately lists Pro and Ultra access in the Gemini app and AI Mode. In other words, app access through Pro should not be mistaken for free API usage.
36
43
Google describes the rollout as global, but that announcement alone does not confirm that every feature is immediately available to every user in every location.
43
API pricing: discounted through November 30
The introductory API rate is $2 per million input tokens and $10 per million output tokens. Cached input tokens cost $0.10 per million, a 95% discount from the input rate. The introductory price is listed through November 30, with rates rising to $4 per million input tokens and $20 per million output tokens on December 1.
1
3
13
The Gemini API supports a unified one-million-token input-and-output limit. That is a shared budget, not one million tokens of input plus a separate million tokens of output.
43
44
For a real cost comparison, estimate the tokens and retries needed for your own task, including cached input and output. Comparing headline rates alone can be misleading: the supplied comparisons do not establish a like-for-like cost per completed task against competing models.
What Google says Argon is built to do
Google positions Argon for complex software engineering, mathematical and scientific problem-solving, autonomous multistep agents, and long outputs. Its enterprise use cases include legal and financial knowledge work, as well as cybersecurity defense. These are Google’s descriptions of the model’s intended strengths, not independent guarantees of performance on a particular team’s workload.
33
43
The practical next step for developers or enterprises is a controlled pilot using representative tasks. Measure completion quality, review effort, reliability and total token use before committing a workflow or budget to the model.
What the benchmark comparisons show
Google’s published benchmark table reports a 68.9% Vals Index score for Argon, compared with 67.0% for Claude Opus 5.5 and 63.1% for GPT-6 Astra. These are results presented by Google; they should be read as benchmark-specific evidence, not proof of broad superiority across tasks.
56
A separate Artificial Analysis comparison reports an Intelligence Index score of 53 for Argon, tied with GPT-6 Astra and below Claude Opus 5.5 at 58 and Claude Sonnet 5.5 at 56.
14
57 The different outcomes illustrate why model choice depends on the task and evaluation—not one headline ranking.
The takeaway
Argon’s public release gives developers a paid API option and brings the model into Google products for eligible subscribers, while its discounted API rates are time-limited. Test it on the work you actually need done, account for the shared token limit and upcoming price change, and treat benchmark leadership as evidence about specific tests rather than a guarantee that it will be the best fit overall.