Kimi K2.7 release date: June 12, 2026, and it is callable today.

Quick answer: the verified date.

Kimi K2.7-Code was released on June 12, 2026. The date is carried by same-day launch coverage (MarkTechPost, June 12, 2026) and matches Moonshot's own announcement channels; the weights went to Hugging Face under a Modified MIT license, and the model went live on the Kimi API and the Kimi Code CLI at release. One naming precision matters for search: the June 2026 release is K2.7-Code, a coding-focused model, not a general K2.7. As of July 2026 Moonshot has not shipped a general-purpose model under the bare K2.7 name; the current generalist remains K2.6. If you arrived here from a config error, the served id is kimi-k2.7-code, with the suffix.

The K2 family history: five releases in eleven months.

Moonshot's release cadence through the K2 line has been unusually fast, and the family tree explains most of the naming confusion in search results. Kimi K2 arrived in July 2025 as an open-weight Mixture-of-Experts model that put Moonshot on the frontier map for coding. K2 Thinking followed in November 2025, the reasoning variant trained to interleave chain-of-thought with tool calls, with Moonshot claiming chains of 200 to 300 sequential tool calls. K2.5 shipped on January 27, 2026, adding vision and the first version of Moonshot's agent-swarm approach. K2.6 landed in April 2026, graduating the coding-preview branch into general availability with claims of roughly 12-hour autonomous sessions and 300-agent swarms. K2.7-Code closed the sequence on June 12, 2026 as a dedicated coding release. All of the recent ids remain served side by side rather than replacing each other, so nothing about the next release changes what you can build on today.

What shipped on June 12: verified vs vendor-graded.

The pricing did not move: Moonshot lists K2.7-Code at the same $0.95 input and $4.00 output per million tokens as K2.6, per its published rates relayed in June 2026 coverage. When a vendor holds price constant and claims lower token consumption, the real-world claim is a lower cost per completed task; that is exactly the number an afternoon A/B on your own tickets settles.

• Verified from the release: open weights on Hugging Face under a Modified MIT license, availability on the Kimi API and Kimi Code CLI, and a 256K-token context window.
• Reported consistently by launch coverage: a Mixture-of-Experts design of about 1 trillion total parameters with roughly 32 billion active per token, the same shape as K2.6.
• Vendor efficiency claim: roughly a third fewer reasoning tokens than K2.6 on comparable coding work. This is a claim you can verify yourself in an afternoon, since reasoning tokens bill as output and the usage log prices both ids on identical prompts.
• Vendor-graded benchmark: a 21.8% improvement over K2.6 on Kimi Code Bench v2, a benchmark Moonshot maintains itself. Independent commentary flagged the self-graded nature at launch, so treat it as directional rather than comparative evidence.

What is not confirmed.

No general-purpose K2.7 and no K3 are announced. Community speculation about a K3 has run ahead of any Moonshot statement, and pages tracking a K3 status list nothing verifiable as of July 2026; the family's fast cadence, five releases in eleven months, invites extrapolation, but a pattern is not a date. The parameter figures above also deserve their hedge: Moonshot's coverage-relayed numbers (1T total, about 32B active) are consistent across sources but are not accompanied by a detailed public technical report for K2.7-Code at the level some peers publish. Nothing in the integration depends on them; they matter mainly for self-hosting math. The practical hedge against roadmap uncertainty is the usual one: integrate against the exact id kimi-k2.7-code behind an endpoint that serves multiple families, and whatever ships next becomes an A/B rather than a migration.

How to call Kimi K2.7 Code today.

The released model is addressable now through any OpenAI-compatible client: base URL https://api.apisrouter.com/v1, model kimi-k2.7-code, one key that also covers kimi-k2.6 for the workloads that are not purely code. No Moonshot account is required through the gateway, and the starting credit means the first requests cost nothing while you evaluate. Both Kimi ids serve a 256K context here, matching the published window for the generation.

import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["APISROUTER_API_KEY"],
    base_url="https://api.apisrouter.com/v1",
)

resp = client.chat.completions.create(
    model="kimi-k2.7-code",   # released June 12, 2026
    messages=[{"role": "user", "content": "What changed between Kimi K2.6 and K2.7-Code?"}],
    max_tokens=2000,
)
print(resp.choices[0].message.content)

Pay-as-you-go · transparent per-model pricing

Selected models are priced below official list prices. Exact input, output, cache, and per-request prices are shown for each model.

Why K2 release dates keep confusing people.

Three patterns generate the noise. First, preview-then-GA staging: K2.6 was confirmed to beta testers in mid-April 2026 and went generally available days later, so coverage written between the two events disagrees on its date. Second, branch naming: the -Code branch shipped as a preview under K2.6 before becoming its own K2.7-Code release, so "K2.7" search results mix the June model with older preview news. Third, platform lag: a model can be live on Moonshot's API days before third-party platforms list it, so launch-week answers to "is it out?" legitimately differ per endpoint. The reliable check beats all of it: list the models your endpoint serves and look for the exact id. If kimi-k2.7-code appears in /v1/models, it is released for you; everything else is press-cycle detail.

curl -s https://api.apisrouter.com/v1/models \
  -H "Authorization: Bearer $APISROUTER_API_KEY" | grep kimi-k2.7

FAQ