Post

just an hour since release, and Kimi K3 is already up on Fireworks openrouter.ai/moonshotai/k... 4 times lower latency, twice as fast looks like they still have some stability problems, but this model will be everywhere in a day or two 🙂 Claude Opus on any plan is probably cheaper now though? 😏

openrouter.aiKimi K3 - API Pricing & BenchmarksKimi K3 is a 2.8T parameter open-weight multimodal reasoning model from Moonshot AI. $3 per million input tokens, $15 per million output tokens. 1,048,576 token context window. Higher uptime with 3 pr...

For the Claude comparison, are you measuring cost per successful result on a fixed workload, with retries and latency included? The $3/$15 rate can look cheap until instability and retries enter the denominator.

no, I realize the list price of Kimi K3 is around half the cost per request on average compared with Opus 5 but on the €18 team subscription, you get way more with a month of daily usage than the list price - from what I've seen, it's a lot more than you'd get for €18 at list price, is it not?

but no coding benchmarks? and most of these models don't even qualify for most tasks? I don't understand how these are scored 🤔

0 reposts0 likes

Qualify depends the threshold %. Say you set it to 90% then the model's score needs to be at least 90% of the model with the highest score, or it does not qualify. No coding. Scope are tasks in enterprise workflows.