Kimi K3 is now live on Morph.

Most open-weight models look promising in benchmarks, then fall apart in production: unreliable tool calls, weak long-context performance, or inconsistent outputs.

GLM-5.2 and Kimi K3 are the first two I’ve tested that feel genuinely production-ready.

In our benchmarks, Kimi K3 approaches Fable-level performance at roughly half the cost of Opus. On Morph, it comes with:

• 1M-token context
• 90–200 output tokens/sec
• Tool calling and structured outputs
• OpenAI- and Anthropic-compatible APIs

We want to see how it performs on real workloads, not toy demos.

If you currently run Anthropic or OpenAI models in production, we’ll cover the cost of routing 5% of your production traffic to Kimi K3. Compare quality, latency, and cost against your existing model with minimal integration work.

We’d love brutally honest feedback: where it works, where it breaks, and what it would take for you to route more traffic to it.

Try it here: https://docs.morphllm.com/sdk/components/fast-models

Comment or DM me if you want us to fund your production test.