Moonshot AI rolled out a HighSpeed version of its Kimi K2.7 Code model on June 15, posting the news to its official account. The pitch is speed: up to 6x faster throughput than the standard release.
The company puts that at around 180 tokens per second on median-length coding inputs, climbing to 260 on shorter-context tasks. Those are Moonshot's own figures. Access is limited for now, which the announcement blames on capacity constraints, and the variant is reaching Kimi Code Beta members, Kimi API developers, and Kimi Business users. No separate invite needed; joining the beta program is enough.
It's the same underlying model as regular K2.7 Code, just tuned for output speed, per the Kimi docs. "Open intelligence should be instant, affordable, and borderless," the announcement says, which is the kind of line that goes in every launch post. Affordable is doing some work there. The post named no prices.
Third-party listings fill that gap. The HighSpeed tier shows up on one model directory at $1.9 per million input tokens and $8 per million output, roughly double the standard model's $0.95 and $4.00. So you pay more for the faster lane. Moonshot hasn't confirmed those numbers itself, and it says the speed may fluctuate while it adds capacity.
Bottom Line
The HighSpeed tier reportedly costs double the standard rate, at about $1.9 input and $8 output per million tokens.
Quick Facts
- Announced June 15, 2026 via Moonshot's X account
- Up to 6x faster throughput (company-reported)
- ~180 tok/s median inputs, up to 260 tok/s short context (company-reported)
- HighSpeed API pricing ~$1.9 input / $8 output per million tokens (third-party listing, unconfirmed by Moonshot)
- Same model as standard Kimi K2.7 Code, 262K context window
