This week, the cost model of AI developer tools changes, a terminal with agents opens its code and another capable open model arrives.
GitHub Copilot moves to token-based billing on June 1
GitHub announced that all Copilot plans move to usage-based billing on June 1, 2026: premium requests give way to AI credits consumed by input, output and cached tokens at each model's API rates. Plan prices do not change, and code completions remain included. According to GitHub, Copilot has become an agent platform with long sessions, and a quick question should not cost the same as hours of autonomous work; a preview bill helps estimate costs before the change.
Companies using Copilot should track consumption by team and set budget limits. The cost of AI in development now depends on how agents are used, not only on the number of seats.
How Dexpert helps: Dedicated Squads & Staff AugmentationSources
- The GitHub Blog: GitHub Copilot is moving to usage-based billing
Warp opens its code and is now built with agents supervised by the community
Warp open-sourced its terminal client and now develops it in a workflow where the community supervises agents from the company's Oz platform, with OpenAI as the repository's founding sponsor. According to Warp, the development bottleneck is no longer writing code but specifying the product and verifying behavior. The company wants to offer an open agentic development environment that works with several models.
The model where people specify and review while agents implement is showing up in real products. Companies can apply the same logic internally by investing in clear specs, rules and tests that make agent-written code reliable.
How Dexpert helps: Artificial IntelligenceSources
- Warp: Warp is now open-source
Mistral Medium 3.5, a 128B open-weight model, runs on just four GPUs
Mistral released Mistral Medium 3.5, a dense 128-billion-parameter model with open weights, a 256k token context and reasoning effort configurable per request. According to the company, it scores 77.6% on SWE-Bench Verified and can be self-hosted on as few as four GPUs; on the API it costs $1.50 per million input tokens and $7.50 output. The model also moves Mistral Vibe coding agents to the cloud, running in parallel and notifying you when they finish.
A strong model that fits in a few GPUs makes it easier to host AI on your own infrastructure, with data under control. It is an alternative for companies with sovereignty requirements or a need for predictable costs.
How Dexpert helps: Artificial IntelligenceSources
Want to put one of these to work in your company?
Dexpert builds custom systems, apps, integrations and AI solutions. Tell us what you need and get an estimate, or take the free digital maturity assessment.