Moonshot Launches Kimi K3 With 2.8T Parameters and 1M Context
Moonshot has launched Kimi K3, making the model available on the web and through APIs. The model has 2.8T parameters and supports up to 1M context tokens.
Moonshot benchmark materials showed Kimi K3 scoring 1687 on GDPval-AA v2, above Claude Opus 4.8 but below GPT-5.6 Sol’s 1,747.8. The company also described the model as built on Kimi Delta Attention and Attention Residuals, while open weights were confirmed for release in the next few days.
From the sources (25 posts)
@teortaxestexDid anyone see it? I can't reproduce Allegedly K3 on July 15th Faster than I'd expect
@synthwavedd🚨 BREAKING: Kimi K3 is launching tomorrow with a discount on top-ups, according to a briefly live article on their docs site
@xeophonRT @synthwavedd: 🚨 BREAKING: Kimi K3 is launching tomorrow with a discount on top-ups, according to a briefly live article on their docs si…
@zephyr_z9LFG!!!!
@kimmonismusLets go: Kimi K3 is launching tomorrow. The discount plan translates into the following: Kimi is launching a limited-time K3 top-up promotion.From July 15 to August 11, users receive bonus credits based on the amount added in a single to
@bdsqlszKimi K3 tomorrow?
@teortaxestexKimi K3 coming soon, LFG! I expect a lot from this model. I'm even forced to admit it will likely mop the floor with V4-official, though uncertainty is really high. All I really hope for is to see the K2 Moment on the next level
@scaling01looks like we are actually getting Kimi-K3 today
@synthwaveddAnother example output from Kimi K3 on @arena. ~Fable level. Damn.
@synthwaveddGive this a peek. Crazy, very excited
@lentils80One of the first Kimi K3 outputs for y'all, tried it on frontend 👀 First impressions is that it's VERY slow, even slower than Fable. This took 35 minutes to finish. However, this is one of the best outputs I've ever seen from this prompt,
@kimmonismusLooks like a big surprise and big release is coming today. Really really looking forward to the K3 release
@teortaxestex? I optimistically predict 53-54 on AA for K3
@lentils80🚨 Kimi K3 is already on LMArena under the stealth name "Kivine" (I know, very stealthy name)
@kimmonismusSuper excited for Kimi K3 (probably) today. 1m context, better long horizon tasks. Lets see how it performs against Opus 4.8/GPT-5.6 (Sol)!
@kimi_moonshot
@mtsliveSITUATION BREWING: Moonshot AI releases a video teasing their upcoming model Kimi K3.
@andrewcurran_It's happening.
@mweinbachRT @Kimi_Moonshot:
@jukan05FT: KIMI TO UNVEIL K3 TONIGHT FT: KIMI’S MODEL IS EXPECTED TO HAVE 2–3 TRILLION PARAMETERS FT: KIMI K3 WILL BE AN OPEN-WEIGHT MODEL FT: KIMI K3 IS EXPECTED TO OUTPERFORM OPUS 4.8, BUT FALL SHORT OF FABLE 5
@jukan05* The reporter said it could be as early as tonight. It’s not confirmed that it will be released tonight.
@synthwavedd🚨 Per Financial Times, Moonshot intend to launch Kimi K3 as early as tonight. K3 is 2-3T total parameters - the largest Chinese model to date - while outperforming 4.8 but staying overall behind Fable in benchmarks as expected
@kimmonismusKimi k3 is being released tonight, via FT -2-3t parameters (Opus4.8 has about 1.5t) -1m context -Expected to exceed Opus 4.8 performance! The time when China was six months behind is over. History is presumably being made today. https://
@scaling01RT @zijing_wu: Chinese AI start-up Moonshot to launch model challenging Anthropic’s lead * Set to release as early as tonight * 2-3T, larg…
@kimmonismusKimi K3 incoming. Most likely today. Btw. banger clip.