Free Intensive Nov 21–22Free 2-day hands-on Vibecoding Intensive on Nov 21–22 at 16:00 MSK (Sat–Sun) · Live Cohort 3 starts Nov 23Join Free →Cohort 3 (Nov 23)
ReleaseCode / SDK · ActionableGitLab Blog→ original

GitLab makes Duo Agent Platform credits last up to 2x longer

GitLab has changed how Duo Agent Platform works: each agent task now needs fewer model requests, so credits should last most customers up to 2x longer. Enabling prompt caching cuts consumption further. Long sessions with large context benefit most, for example when an agent takes a task all the way to a merge request. So before December 18, 2026, it's worth comparing your own usage and deciding whether to move to the new terms yourself.

AI-processed from GitLab Blog; edited by Hamidun News
GitLab makes Duo Agent Platform credits last up to 2x longer
Source: GitLab Blog. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceGitLab Blog ↗
Editor-in-Chief
Key metric #1New terms for new customers apply from November 19, 2026
Key metric #2Customers who opt out before December 18 keep the old terms until renewal or for 12 months
From January 1, 2027, a per-seat bonus$8 per month for Premium and $16 for Ultimate

The biggest headache for companies deploying AI agents is an unpredictable bill. An agent picks up a task, calls a language model dozens of times, and credits vanish faster than expected. GitLab has announced that Duo Agent Platform is now more economical: the same GitLab Credits should last most customers up to twice as long.

What exactly changed

The platform now uses the models GitLab offers inside the product more efficiently. These include recent models from Anthropic, OpenAI and Google, as well as open-weight models hosted by GitLab itself. Each task you hand to agents and flows (ready-made automation scenarios) now generates fewer model requests. Credits are charged per request, and each type of model has its own exchange rate. The rates have not changed, so fewer requests means fewer credits.

There is a second lever: prompt caching. With it enabled, eligible tasks use even fewer requests. The simplest way to picture it is that the agent doesn't have to "re-read" the same large chunk of context every time. Long sessions with heavy context benefit most. GitLab names an example directly: Duo Developer, which takes an issue and carries it through to a merge request that is ready for review.

As an illustration, the company cites a real 500-person team in October 2026: for every scenario, the number of billed requests is higher than it would have been with the new optimizations. Specific amounts are not disclosed in the text; GitLab suggests reviewing them with your GitLab account team.

An important caveat: "up to 2x" is a maximum, not a guarantee. For some customers the savings will be more modest. In addition, fixed-price features (Code Review Flow, SAST flows and Code Suggestions) are not affected by the changes, and self-hosted models stay at 8 requests per credit.

Where GitLab suggests spending the savings

The logic is straightforward: if scenarios cost less, you can run more of them. The company lists several options:

  • the agent in GitLab for Slack reads a discussion in a channel, turns decisions into issues and merge requests, and posts the result back to the thread;
  • scheduled automations (beta in GitLab 19.6) run flows on a schedule, from every 15 minutes to once every three months, for example for morning issue triage or CI health checks;
  • Business Logic Security Analyzer (beta for Ultimate on GitLab.com) looks for authorization and access-control vulnerabilities that pattern-based scanners miss; to run it on Claude Mythos 5.2, you need to contact your account manager;
  • the /goal command in Duo CLI and chat: you describe a goal, and the agent plans, executes and verifies the result, with an independent "judge" evaluating the outcome. You can pause or redirect the work.

For customers with a GitLab Flex contract, freed-up credits can also go toward other products: Secrets Manager, Dedicated Runners and, soon, GitLab Orbit.

Spending controls and transition deadlines

GitLab also points to its usage-tracking tools. The Impact dashboard (since version 19.5) shows consumption by model, flow type, group and project. The Credit caps page (since 19.4) lets you set a per-user credit limit and individual exceptions without the API. Regular users on GitLab.com can now see their own usage.

The timeline is as follows. For new customers, the new terms apply from November 19, 2026. Existing customers can switch them on at any time from that date, but there is no going back. If you do nothing, the transition happens automatically on December 18, 2026. Customers who opt out before that date keep their current terms and included credits until the next renewal or for 12 months, whichever comes first.

Separately, the promotional per-seat bonus is changing: from January 1, 2027, it will be $8 in credits per month for Premium and $16 for Ultimate.

What to do in practice

For anyone already paying for agents in GitLab, this is an almost free improvement, but don't take it on faith. Ask your account team to recalculate your recent usage under the new rules, and enable prompt caching. This applies especially to teams whose agents run long sessions, such as Duo Developer. If your budget currently rests on the old terms, weigh whether you need the deferral: because opting in is one-way, it makes sense when the recalculation shows real savings. Also keep in mind that this is the vendor's own announcement, and it includes no independent measurements.

Story

Key Questions Answered

What exactly changed?

The platform now uses the models GitLab offers inside the product more efficiently. These include recent models from Anthropic, OpenAI and Google, as well as open-weight models hosted by GitLab itself. Each task you hand to agents and flows (ready-made automation scenarios) now generates fewer model requests. Credits are charged per request, and each type of model has its own exchange rate. The rates have not changed, so fewer requests means fewer credits. There is a second lever: prompt caching.

Where GitLab suggests spending the savings?

The logic is straightforward: if scenarios cost less, you can run more of them. The company lists several options: the agent in GitLab for Slack reads a discussion in a channel, turns decisions into issues and merge requests, and posts the result back to the thread; scheduled automations (beta in GitLab 19.6) run flows on a schedule, from every 15 minutes to once every three months, for example for morning issue triage or CI health checks; Business Logic Security Analyzer (beta for Ultimate on…

What are the key details on Spending controls and transition deadlines?

GitLab also points to its usage-tracking tools. The Impact dashboard (since version 19.5) shows consumption by model, flow type, group and project. The Credit caps page (since 19.4) lets you set a per-user credit limit and individual exceptions without the API. Regular users on GitLab.com can now see their own usage. The timeline is as follows. For new customers, the new terms apply from November 19, 2026. Existing customers can switch them on at any time from that date, but there is no going back.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Unified AI Workspace · Nomad

Test Frontier LLMs & Autonomous Agents in One Workspace

Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.

Try Nomad AI →Launch Nomad Agent
What do you think?
Loading comments…