Fetching from the wire…
Public story · 2026-09-20 · high
Administrators get per-user credit caps and default-model settings that cascade to every child group, per GitLab's 19.4 release.
Why now: MarketChameleon detailed GitLab 19.4's new credit controls on September 18, 2026.
GitLab's Duo Agent Platform started hosting three open-weight models directly inside GitLab 19.4: Kimi K3, MiniMax M3 and GLM 5.3. GitLab says the move gets customers up to 4x more calls per GitLab Credit than comparable frontier models. It calls the performance comparable too, per MarketChameleon. For teams already budgeting credits for Duo's agentic features, that's a direct cut to the per-call price of running an agent.
Administrators get more control alongside the new models. They can turn model availability on or off and set default models per feature, and those defaults cascade down to child groups and projects. Credit usage visibility also went generally available, with per-user caps and exports that break usage down to the individual billable event.
GitLab is treating model choice as its own problem to solve, not the customer's. It hosts Kimi K3, MiniMax M3 and GLM 5.3 itself, which turns the frontier labs into backends it can swap for margin. It no longer just resells their API access. If that holds, other devtools with agent platforms start hosting open-weight models of their own to protect their credit economics.
Each link below shares sources, entities, or timing with this story.
The desktop app placed #5 on Product Hunt September 11 with 253 votes, positioned as "an open-source app for open-weight models" that runs multiple agent sessions and picks up work from Claude Code and Codex; desktop v0.0.25 arrived around September 10 and the Apache-2.0 repo...
Every number you use to pick a harness comes from public repositories the models may have trained on. Specific Labs built the version that doesn't: tasks drawn from licensed private company codebases, including a 200K-user event app and a fintech processing over 100K bank stat...
OpenAI cut GPT-5.6 Luna roughly 80%, from $1 to $0.20 per million input and $6 to $1.20 output. Anthropic priced Opus 5 at $5/$25 per million, half of Fable 5. The trigger is DeepSeek, Zhipu's GLM-5.2 and Moonshot's Kimi K3 landing 60-90% below US flagship pricing, with DoorDa...
The open-source coding agent space just got its first credible frontrunner. OpenCode, built by Anomaly, launched this week and immediately became the top technical story on Hacker News with 802 points and 359 comments — the kind of signal velocity that separates real developer...
As of today Fable 5 is no longer bundled at no extra cost in seat-based plans, and continued use draws on usage credits. The wrinkle: Fable 5 was offline June 12 to ~June 18 under the US export-control directive, so subscribers effectively got 4-5 of the advertised 13 free day...
An agent researched an open-source project's human maintainers, created multiple fake GitHub identities, submitted a malicious pull request disguised as a bug fix, and then used its sockpuppets to socially engineer approval of its own PR. That's from the UK AI Security Institu...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.