Is Kimi K3 Available on EvoLink? Expected July 17, 2026
kimi-k3 identifier, or promise K3 access to customers.Kimi K3 status as of July 16, 2026
| Question | Current status | What developers should do |
|---|---|---|
| Has Moonshot released Kimi K3? | Yes | Treat K3 as an official upstream model, not a rumor |
| Is Moonshot's direct API live? | Yes | Direct Moonshot customers can evaluate the documented kimi-k3 route |
| Is Kimi K3 available on EvoLink? | Not yet | Do not send production requests using an assumed EvoLink route |
| Is there an EvoLink model ID? | Not published | Do not assume it will match Moonshot's direct ID |
| Is EvoLink pricing available? | Not published | Do not convert Moonshot's RMB prices into an EvoLink price claim |
| Is there an EvoLink launch date? | Expected July 17, 2026 | Follow this article and the official model page |
| Should teams prepare evaluations? | Yes | Build replayable tasks, budgets, and fallback policies now |
What Moonshot has officially confirmed
Moonshot's official documentation lists:
- direct API model ID
kimi-k3; - 2.8 trillion parameters;
- Kimi Delta Attention and Attention Residuals;
- a 1,048,576-token context window;
- text, image, and video input;
- always-on reasoning;
reasoning_effortwithmaxas the current supported level;- tool calling,
tool_choice, dynamic tool loading, structured output, and automatic context caching; - direct prices of ¥2/MTok cached input, ¥20/MTok uncached input, and ¥100/MTok output.
These are Moonshot facts. They should not be rewritten as EvoLink route specifications until EvoLink verifies the integration.
Why gateway integration needs separate verification
An OpenAI-compatible upstream API reduces integration work, but it does not remove the need for route testing.
| Verification area | Why it matters for K3 |
|---|---|
| Model naming | EvoLink must publish the exact supported route ID |
| Reasoning fields | Multi-turn requests need complete assistant messages and reasoning_content handling |
| Parameter compatibility | K3 uses reasoning_effort, not the K2.x thinking control |
| 1M context | The gateway must accept, route, meter, and time out very large requests correctly |
| Streaming | Reasoning and final-answer deltas must be represented consistently |
| Tool calls | Tool IDs, arguments, dynamic definitions, and follow-up messages need compatibility tests |
| Structured output | JSON Schema behavior should be checked against the gateway response |
| Multimodal input | Image and video upload or URL handling may differ from Moonshot's direct interface |
| Billing and caching | Cache hits, input/output usage, retries, and failed requests need reliable accounting |
| Capacity and fallback | Launch demand can expose latency, 429, 5xx, and timeout behavior |
Publishing a route before these checks would create a worse developer experience than waiting for a verified integration.
What developers should use now
The useful preparation work is provider-independent:
- Put model choice behind configuration.
- Keep a current production route as the baseline.
- Build a 20–50 task K3 evaluation set using real coding, research, multimodal, and tool workflows.
- Define acceptable latency, cost, retry rate, and reviewer effort.
- Decide whether K3 would be a default, specialist, premium, or escalation route.
- Preserve a fallback for provider errors and capacity limits.
What will change when EvoLink support is verified
This URL is designed to remain useful through the integration lifecycle.
When K3 becomes available on EvoLink, this article should be updated with:
- the exact launch date;
- a link to the canonical Kimi K3 model page;
- the verified EvoLink model ID;
- current EvoLink pricing through the existing pricing system;
- supported request interfaces and modalities;
- context and output limits;
- migration notes and known compatibility differences.
Kimi K3 API, Kimi K3 pricing, and EvoLink model-ID intent. This article will remain as the integration-status and launch-history record.What we will not claim before verification
Until a live EvoLink route and documentation exist, this article will not claim:
- that Kimi K3 is available through EvoLink;
- that Moonshot's direct model ID works on EvoLink;
- that Moonshot direct pricing equals EvoLink pricing;
- that all 1M tokens are supported through the gateway;
- that every Moonshot tool, visual, caching, or streaming behavior is identical;
- that failed requests have a particular billing policy;
- that the route is production-ready in every region.
That fact boundary protects developers from building against a route that does not exist yet.
FAQ
Is Kimi K3 available on EvoLink?
Not yet. EvoLink has not published a verified K3 model page or API route as of July 16, 2026.
Is Kimi K3 officially released?
Yes. Moonshot published K3 in its official model list, quickstart, pricing page, and direct API examples on July 16, 2026.
Can I use kimi-k3 as the EvoLink model ID?
kimi-k3 is Moonshot's documented direct API ID. Wait for EvoLink to publish its verified route name.How much will Kimi K3 cost on EvoLink?
EvoLink pricing has not been published. Moonshot's direct price cannot be presented as the future EvoLink price.
When will EvoLink add Kimi K3?
EvoLink expects to launch Kimi K3 on July 17, 2026. The final model ID, price, and supported capabilities will be confirmed on the official model page and API documentation.
Why not expose the route immediately if Moonshot is OpenAI-compatible?
Compatibility still needs verification across reasoning fields, tool calls, streaming, multimodal inputs, 1M context handling, billing, caching, timeouts, and fallback behavior.
What should I use while waiting?
Should I test Kimi K3 through Moonshot directly?
That depends on your procurement, data, regional, and account requirements. If you test it directly, keep the evaluation portable so the same workload can be repeated against a future EvoLink route.
