Gemini 4 Argon lands on GitLab Duo Agent Platform for long-running tasks
GitLab has made Google's Gemini 4 Argon available on its Duo Agent Platform, positioning the model for long, multi-step agentic work across the software lifecycle. According to GitLab, the model completed 93.3% of the tasks in its internal software engineering benchmark.
- Google's Gemini 4 Argon is now available on GitLab Duo Agent Platform.
- GitLab says Argon completed 93.3% of its internal software engineering benchmark tasks and resolved 73.3%.
- Of runs that hit a tool error or timeout, Argon still completed 93.8%, per GitLab.
- Promotional pricing is 2.86 calls per credit through November 30, 2026, then 1.33 calls per credit from December 1.
- Flash models remain the lower-cost option for high-volume work.
Built for long-horizon work
GitLab says the challenge with AI models handling longer, multi-step work is applying that capability across the software lifecycle without adding risk or cost. Teams, it argues, need to move faster within enterprise controls and guardrails. Gemini 4 Argon, now on GitLab Duo Agent Platform, was built for that kind of work, according to the company.
GitLab says Argon sustains its reasoning across the many steps of a task, including reading files, running commands and editing code through to completion. The idea is that engineers can delegate work to GitLab Duo agents powered by Argon and spend their time reviewing results instead of monitoring runs. Agent output lands in a merge request, where the team can review the change with its full context.
Benchmark results and safeguards
GitLab reports several figures from its internal evaluations:
- Argon completed 93.3% of GitLab's software engineering benchmark tasks.
- It resolved 73.3% of the benchmark tasks, with each change passing the benchmark's tests across multiple programming languages.
- Of the runs that hit a tool error or timeout, Argon still completed 93.8%, so a single mid-run failure should not force a team to start over.
These numbers come from GitLab's own testing and have not been independently verified in the source.
On behavior, GitLab says Google trained Argon to respect system boundaries, to tell users when it could not find something or finish a task, and to pause for confirmation before an irreversible step. GitLab describes Argon as a strong choice when finishing correctly matters more than finishing fast.
Where Argon fits among Gemini models
GitLab calls Gemini 4 Argon Google's newest frontier model and the most performant Gemini model on Duo Agent Platform. It handles long-horizon agentic actions especially well, the company says, including tasks that begin loosely defined. It sits alongside the Flash models, which GitLab describes as a lower-cost option for high-volume work, so teams can match each workload to a suitable Gemini model.
Whichever Gemini model is chosen, it runs inside the same context layer, policy checks and audit trail that cover every agentic action across the software lifecycle, according to GitLab.
Pricing and how to try it
Argon runs on GitLab Credits, like other models on the platform. Through November 30, 2026, it is offered at promotional pricing of 2.86 calls per credit. Starting December 1, standard pricing of 1.33 calls per credit applies.
Existing Duo Agent Platform users can select Gemini 4 Argon in model selection. GitLab Premium and Ultimate customers can turn on Duo Agent Platform and try Argon using the GitLab Credits included with their subscription. Teams new to the platform can start a free trial.
For customers, the pricing change means the same credit buys fewer Argon calls after the promotion ends, a factor to weigh against the cheaper Flash models for high-volume work.
