Announced 30 Sept 2026 · Sources checked
What did Google announce?
Google positions Argon for extended coding, professional knowledge work, and defensive security tasks. It reports a one-million-token output limit. The introductory API rates are $2 per million input tokens and $10 per million output tokens; Google lists later rates of $4 and $20 respectively.
The broader rollout is planned to start with paid API customers and Google AI Ultra subscribers. The announcement gives no universal public-access date.
What should a builder compare first?
A useful comparison begins with a task your team already understands. For a hypothetical coding pilot, use an old bug with a known resolution, a separate feature request, and a repository the assistant has not seen. Give each candidate the same context and the same acceptance criteria.
Keep the final patch, failed attempts, reviewer corrections, and total usage together. A polished explanation can conceal an incorrect change. A benchmark result can help you choose candidates, but it does not replace inspecting what happened in your repository.
Why does a longer run change the review?
Longer workflows create more intermediate decisions to inspect. Ask where the model can write files, which systems its tools can reach, and when it must stop for human approval. A successful small task does not establish that the same setup can safely execute a much larger one.
For example, a trial agent could prepare a proposed dependency update and a test report while a maintainer decides whether to merge. Track mistaken edits and unnecessary changes as well as passing tests. That makes the review about the actual work rather than the confidence of the response.
What remains uncertain?
The practical decision is still workload-specific. This publication has not run an Argon comparison. Before adopting it, verify access in your own account, current commercial terms, and results on representative tasks. Keep a working fallback during any trial.
Common questions
Does this mean anyone can use Argon now?
No. The cited announcement describes phased access.
Is this an independent model review?
No. It separates Google’s announcement from an original evaluation checklist.
What to remember
Google has announced its next frontier model, with early access focused on trusted cyber defenders. Here is what builders can assess now.
Sources & further reading
How this story was made
Written by Kristian Kostov with AI assistance and checked against the linked sources. Company performance claims are attributed to the company. Analysis reflects AiLookout’s interpretation; we have not independently tested the products discussed. Cover photography is illustrative and does not depict the specific announcement or product.
Our editorial standards
