Edgee Turbo Models - Use Claude Code with Kimi K2.7 Code, MiniMax M2.7, and more

Haha, Thank you @denis_lt , that means a lot.

Kimi K2.7 Code on Turbo is genuinely a sweet spot right now. The combination of the model's coding capability and the speed/price of how it's served changes the calculus for a lot of agentic workflows.

Keep building 🙏

@thenameisarian Our defaults target ~50% reduction with measurable accuracy loss

in the noise on coding benchmarks. Past ~70% you start seeing degradation on long-horizon tasks (multi-step refactors, anything that needs to recall earlier tool results). Single-step tasks

tolerate more.

It's less "how much can I compress" and more "what can I drop without breaking the loop".
We have implemented three compression techniques:
- Tool result trimming by RTK
- Tool surface reduction
- Output brevity

We are trying to make each of these techniques as lossless as possible, and I can already say that the effects on the model's efficiency are almost negligible. We will be releasing our benchmark work within the next few days.

@chaseforbis98 thanks for your questions, let me clarify a few things:

Turbo Models is a feature inside Edgee, not a single model. It lets you run frontier open-source models (GLM 5.1, Kimi K2.7 Code, Kimi K2.6, MiniMax 2.7) directly inside Claude Code or Codex. So you're not building apps "with Turbo", you're using Turbo to power your coding assistant when you build whatever apps you want.

Yes, flat $29/month per developer. Not $29.99. The plan includes a generous monthly usage allowance that covers full-time intensive coding for the vast majority of developers. There is a ceiling at the very high end (for context, you'd need to be running agentic loops nearly continuously to hit it), and if you ever get close, we'll talk before anything changes. Transparent and fair.

On "no more token manipulation": you're right that you stop worrying about the per-token meter, which is the big mental shift. But Edgee actually does smart token compression behind the scenes (cutting what gets sent to the model by ~50% on coding sessions), so you get the benefit of token optimization without having to think about it. Set it and forget it.

Hope this clarifies. Happy to go deeper on anything 🙏

@crystalmei Thanks 🙏

Three layers handle this:

1/ Gateway: requests don't queue locally. Per-provider HTTP/2 connection pools with multiplexing, horizontal scaling, no shared serialization point. A burst from one team doesn't slow down another.

2/ Inference: Turbo runs on dedicated high-throughput infra with our partners ( @Together AI , @Fireworks - Fastest Inference for Generative AI , others depending on model). They auto-scale and load-balance across replicas. We've sized partnerships with headroom for peak hours.

3/ Fallback: if a Turbo lane is ever saturated, Edgee routes automatically to a standard endpoint of the same model family. The agent loop never breaks, the user might just see a slightly slower response for that call.

Plus per-key rate limiting at the team level so one heavy user doesn't degrade the experience for the others.

🙏

@olivier_thirion_de_briel Thanks for being one of our core users, it means a lot to us. And yeah, your savings are incredible. Can't wait for your to test our next compressor ;)