Skip to content
MustHave.ai Musthave.ai
Models

Gemini 4 Argon Is Now Available: API Access, Pricing and Output Limits

Gemini 4 Argon API access, pricing and output limits, plus a downloadable 18-check coding fixture pack with an explicit testing method.

Gemini 4 Argon Is Now Available: API Access, Pricing and Output Limits
The standard Argon input and output rates double on December 1, 2026.

November 5 update: a reproducible coding test pack

Before comparing Gemini 4 Argon with another coding model, give both the same task and acceptance tests. We prepared a downloadable Python fixture pack with three contracts: parse dollar amounts exactly, update a vote score correctly, and merge product pages without duplicate IDs.

The pack contains a prompt, an 18-check harness, and a hand-written reference implementation. Our local run passed all 18 reference checks. A deliberately broken baseline failed 9 of 18. Those runs validate the fixtures against known mistakes; they do not measure Gemini.

No Gemini API output was generated or scored for this update. The tests are open book, with acceptance cases included in the download. They cannot support a blind-benchmark claim or a general model ranking.

Keep failed attempts in the comparison

Ask each model to produce candidate.py from the included prompt. Review the file before running it. Use a disposable environment without secrets, network access, or valuable files, because the harness imports and runs the submitted module.

Record the exact model identifier, provider, and thinking setting, along with the output limit and elapsed time. Save token usage, the applicable rate card, raw output, and every retry. A repaired solution is a new attempt; keep its first failure in the record.

These small checks target common application errors. A floating-point shortcut can mishandle money. Replacing a downvote with an upvote needs a two-point score change. Merging feed pages must retain the first occurrence without altering the input records. Passing them does not establish repository-scale engineering ability.

Recheck the rate before a billed run

Google’s current standard API pricing still lists $2 per million input tokens and $10 per million output tokens through November 30, 2026. From December 1, those rates become $4 and $20. Thinking tokens are included in billed output. The model is not available on the Free tier.

The model reference continues to identify gemini-4-argon as the stable model. The earlier access and output-limit coverage follows; this update adds a test method without claiming new release availability.

Argon now has a documented API endpoint. Before moving a workload, check the access tier, output mode, and rate that apply to the request.

October 15 update: Argon in Copilot and Figma Make

GitHub’s October 13 release makes Argon generally available to Copilot Pro+, Max, Business, and Enterprise users. Rollout is gradual across supported clients. Business and Enterprise access follows model policy, and GitHub bills the model at provider list pricing under usage-based billing.

Figma’s October 13 release notes also list Argon rolling out in Figma Make on all plans. A platform model picker and direct API access have different billing contexts. Check the surface you will use before applying the API price calculation below.

For the surrounding design changes, refer to our Figma code layer and backend guide. Record the client, plan, and enabled model policy with a trial result; availability in one surface does not establish access in another.

October 13 replaces the earlier access plan

Google’s API changelog records general availability on October 13, 2026. The stable model identifier is gemini-4-argon. Our original October 1 report described a future rollout; the documentation now establishes public API availability.

The current price card excludes the Free tier. Confirm API billing eligibility and quotas for your project; a consumer subscription and a billed developer project have different access contexts.

Use the effective date in a cost estimate

Standard API input costs $2 per million tokens through November 30, 2026, then $4 per million tokens from December 1. Output, including thinking tokens, costs $10 per million tokens through November 30, then $20. Batch, Flex, and Priority have separate rate cards.

For an illustrative standard request with 100,000 uncached input tokens and 10,000 billed output tokens, the token charge is $0.30 before December. It becomes $0.60 at the December rates. This example excludes grounding, retries, and other billed services.

Default output and long decode differ

The model reference lists a unified input-and-output context limit of 1,048,576 tokens. Default synchronous output is capped at 65,536 tokens. Long decode supports up to 1,048,576 output tokens within the documented context boundary.

A large output ceiling is not a sensible default budget. Set a lower limit that fits the task, then inspect the resulting artifact. Include reasoning output in the spending estimate because the price card includes thinking tokens in the billed output category.

Evaluate the endpoint with one reproducible task

  • Record the exact model identifier and request mode.
  • Save the price card and effective date.
  • Set token, time, and total-cost limits.
  • Use non-sensitive material for the first request.
  • Check the output against written acceptance criteria.

Start with a task that already has a known answer or a testable patch. Keep failed attempts in the cost record. General availability establishes a release state; it does not establish performance on your project.

Our work-agent permissions guide and Postman evaluation guide cover related trial boundaries.

Budget for the rate after the introduction ends

Choose a workload that remains affordable at the December rate. What accepted result would justify using Argon instead of the endpoint you already run?

Abdessalam Alaoui

I'm Abdessalam Alaoui, founder of Musthave.AI. I write practical guides to AI models, tools, generative media.

Join the discussion

Your email address will not be published. Required fields are marked *

Keep reading

All news