MODEL PROFILE · Google DeepMind
Gemini 4 Argon
A named model with restricted initial access. Separate its announced token prices and output limit from public endpoint availability.
Source reviewed: · Language model · restricted rollout
Provider documentation summary. No hands-on results or independent ranking are claimed.
Documented facts
The following fields come from the official provider source, reviewed on the date above.
- Identity and announcement
- Google DeepMind announced Gemini 4 Argon on September 30, 2026. A public API identifier is not established here.
- Access
- Initial rollout through Fairwind to trusted cyber defenders. Broader release is forthcoming, starting with paid API customers and Google AI Ultra subscribers; no date is specified.
- Provider-described uses
- Long-running software engineering, enterprise knowledge work and defensive cybersecurity; multimodal understanding is described. Exact API input formats remain unverified.
- Output limit
- Google states 1 million output tokens. This is not a verified input or total context-window limit.
- Announced introductory pricing
- USD $2 input and $10 output per million tokens; cached input is 95% cheaper, implying $0.10 per million cached input tokens.
- After introductory pricing
- USD $4 input and $20 output per million tokens. Introductory expiry date, other charges and subsequent cached-input terms are not established.
- Weights and license
- No downloadable weights or open-weight license established in this review.
The practical decision
The immediate decision is whether access fits your evaluation schedule. An announced model is not automatically an endpoint your account can invoke. Keep a working language model for delivery commitments and reserve an Argon comparison until you can record the exact service, version and settings. Treat provider capability descriptions as hypotheses to evaluate on your own tasks, not an independent recommendation.
What to record in your own comparison
- For coding, prepare reproducible failing cases and regression checks. Require an inspectable change, not merely a persuasive explanation of a proposed fix.
- For document work, preserve source material and assess whether conclusions can be traced to it. Measure omissions and unsupported assertions alongside task completion.
- Budget input and output separately. As arithmetic only, 100,000 uncached input tokens plus 10,000 output tokens would cost $0.30 at the announced introductory rates or $0.60 at the stated later rates, excluding other charges. This is not a usage measurement or a current access offer.
- Keep the output ceiling separate from the amount a task needs. Set practical stopping criteria and record retries before comparing cost per completed task with an available alternative.
Limitations and unknowns
Documentation research only; no model requests, performance measurements or independent results. Context capacity, precise modality schemas, endpoint ID, regional eligibility and general availability remain unverified. The earlier anonymous Arena reports do not establish Argon’s identity in those sessions.
Compare your saved results
Use the DualView editor to inspect files you already have. The accepted-output calculator uses your own rates and counts. Neither link runs a model or purchases generation.
Related reporting
- Gemini Arena rumors: announcement follow-up and remaining uncertainty
- GPT-6.1 Sol: a separate language-model profile for evaluation planning
- Claude Opus 5.5: separately documented model profile