GitLab adds open-weight models, caps and analytics to curb AI agent spend

Matching the model to the task

According to GitLab, agent cost starts with the price of each model call. Duo Agent Platform lets admins pin a default model per feature and decide what teams can use: their own model, self-hosted or from a provider, GitLab-managed frontier models, or the GitLab-managed open-weight models that are new in GitLab 19.4.

Three open-weight models are now hosted in the platform: GLM 5.3, Kimi K3 and MiniMax M3. In GitLab's internal benchmark with Duo agents, the company says GLM 5.3 matched Claude Opus 5's task completion at approximately one-fifth the cost. Kimi K3 matched Claude Fable 5.1 at one-third the cost, and MiniMax M3 matched GPT-5.5 at one-eighth the cost.

Model choice is an admin setting. In the top-level group's GitLab Duo settings, admins open Model Selection, select Manage Models and pin a model to each feature. GitLab's example is pinning Kimi K3 to Agentic Chat and MiniMax M3 to agents and flows, so routine work does not default to a frontier model. Group Owners also choose which models their teams can use.

Fewer retries with Orbit

Model choice sets the price of each call, while retries multiply the number of calls. GitLab cites a Gartner finding that a team of 40 developers spent 23 times more on AI because agents kept retrying without the right rules and context.

GitLab Orbit maps code, work items, merge requests, pipelines, deployments and security findings into a single graph that stays current as teams ship. Agents in Duo Agent Platform query it natively. External agents such as Claude Code and Codex connect through Model Context Protocol (MCP), and custom tooling uses the REST API. Every agent sees exactly what its user can see in GitLab.

In GitLab testing, the same task in Claude Code needed up to 45x fewer retries with Orbit and, separately, used 4.5x fewer tokens to get the right context the first time. Orbit is in free beta on GitLab.com and, according to GitLab, reaches general availability next month across all deployment options.

Visibility and spending limits

Duo Agent Platform Impact Analytics breaks spend down by team, task and model for a chosen set of teams and date range. It also shows how an open-weight model compares with the current model mix. The GitLab Credits dashboard goes below the team level, tracing usage to the user and session behind it. Impact Analytics also shows credit consumption alongside the work that reaches production.

Credit and usage controls set caps at the subscription, group or user level. They send alerts at 50%, 80% and 100% of spend and pause usage automatically once the cap is reached. GitLab gives the example of a flat cap of 1,000 credits per user per month, with an individual override such as 500 additional credits for an engineer who is experimenting.

Per-user caps apply once a user has consumed their included credits. Only the capped user loses access to credit-consuming features, until the next billing period or until an administrator raises the cap.

Availability and where to start

GitLab-managed open-weight models and credit and usage controls are generally available. Duo Agent Platform Impact Analytics is in early access. GitLab suggests starting with the model by pinning GLM 5.3, Kimi K3 or MiniMax M3 to a routine workload, such as a flow that runs on every merge request. Next, turn on Orbit for the group that runs the most agent sessions, then set per-user caps with overrides for heavy users. After a few weeks, teams can compare usage by team and model against the previous period.

New users can start a free trial of GitLab Duo Agent Platform. Premium and Ultimate customers can turn it on and use the GitLab Credits included with their subscription. The cost claims come from GitLab's own benchmark and testing, so results may vary by workload.