Gemini 4 Argon: General availability, per-token pricing

1. Model availability and performance claims

GitHub announced general availability of Gemini 4 Argon on October 13, 2026. The model targets complex coding, reasoning, and agent use cases involving multiple tasks.

According to the announcement, early tests showed it solved more tasks than previous Gemini models across all coding benchmarks tested, with the largest gains in Copilot CLI. No benchmark scores or test conditions were provided, making quantitative comparisons of the gains difficult.

GitHub says the model checks its results before completing a task and tests edge cases beyond typical inputs.

2. Eligibility and rollout

For individuals, eligible plans are Copilot Pro+ and Copilot Max. Users can select the new model from the model picker.

For organizations, eligible plans are Copilot Business and Copilot Enterprise. Administrators can manage access through model policies in settings. The new model is allowed automatically unless the global default is turned off or the model is explicitly disabled.

Supported environments include development editors, command-line tools, coding agents, and apps. For example, users can select it in Visual Studio Code and Copilot CLI.

The rollout is phased. Even with an eligible plan, the model may not yet appear in the picker, and GitHub advises checking again later.

3. How charges are calculated

Usage-based billing counts content sent to the model as input tokens and generated content as output tokens. Tokens measure the content exchanged with the model, while cached tokens represent reused or stored context. Charges calculated at model-specific rates are converted into AI credits, with 1 credit worth $0.01.

Individual plans include plan-specific credit allowances. Organization plans pool per-user credits at the billing-entity level. Usage beyond the included allowance incurs additional charges at the token rates.

If you subscribe or have ever subscribed through GitHub Mobile, you cannot purchase additional credits.

Separate model multipliers apply to existing annual Copilot Pro and Copilot Pro+ subscribers on request-based billing. The pricing documentation distinguishes this system from usage-based billing.

4. Gemini 4 Argon rates and scope

The pricing documentation states that all prices are per 1 million tokens. Gemini 4 Argon costs $2 for input, $0.10 for cached input, and $10 for output. These promotional rates apply through December 1, 2026; the documentation does not list subsequent prices.

Having another option for complex tasks is a benefit. However, usage-based billing requires budgeting for input and output volume, not just task counts. Since this model charges different rates for output and input even for the same 1 million tokens, input volume alone is insufficient to estimate costs.

5. Billing boundaries by feature

Code completion and next edit suggestions do not consume AI credits and are unlimited on paid Copilot plans. The new model's token rates should not be applied directly to these features.

Code review, by contrast, automatically selects its model without disclosing it, so token rates may vary by review. Costs arise from both token usage and GitHub Actions runtime, making them difficult to predict using this new model's rates alone.

Assessing the practical value of this general release requires considering not only model choice but also organizational access policies and usage-based budgets.