Seriously, this is such an insane release, I did not expect this:
Kimi K3.1 Pro beats every listed US frontier model on 6 of 14 benchmarks in Moonshot’s evaluations.
At $2 per million input tokens and $10 per million output tokens, its standard token rates are 50% below Opus below GPT-6 Astra ($10/$50!!)
In other words: SOTA much cheaper, local running (with enough compute)
Kimi K3.1 Pro beats every listed US frontier model on 6 of 14 benchmarks in Moonshot’s evaluations.
At $2 per million input tokens and $10 per million output tokens, its standard token rates are 50% below Opus below GPT-6 Astra ($10/$50!!)
In other words: SOTA much cheaper, local running (with enough compute)
Holy shit: Kimi k3.1 pro is on par or even outperforms Opus 5.5 and GPT-6 astra on key benchmarks!!
They just announced K3.1 Pro, a 2.8T-parameter multimodal model with a 1M-token context window that it claims rivals leading proprietary models in coding and agent tasks, with y November 4.
China is back!!
Oct 23, 2026 · 5:32 PM UTC
133
173
2,805
190,256
@kimmonismus How is it cheaper? Nobody is paying API rates for Claude, and the included usage for the kimi plan is hilariously bad. Opus 5.5 is, in reality, way way cheaper to actually use.
5
0
91
5,172
@kimmonismus this graph is stupid. it compares opus 5.5 and sonnet 5.5 are what effort? just a dot. they are comparing kimi 3.1 pro max that beats opus 5.5 ??? low?
1
0
19
2,572
@kimmonismus I’m tired of the incremental rule labs apply to their release. 3 months without release, ship a monster, crush all benches, +0.1….
Bro
Bro
2
0
20
3,815
@kimmonismus does anyone actually care about benchmarks anymore? i cant wait until the day where a model is just released. no benchmarks, nothing. just like.. hey guys new model. check it out.
1
0
8
2,884
@kimmonismus @AnthropicAI about to start complaining about distillation and model routing lmaoooo
People forget, yes, China doesn't have the best chips, but it does have wayyy too much energy, and it can achieve the same performance by running 10 chips instead of 1
People forget, yes, China doesn't have the best chips, but it does have wayyy too much energy, and it can achieve the same performance by running 10 chips instead of 1
7
0
20
3,238
@kimmonismus these KIMI benches are SCAM, the used claude code harness for them
0
0
1
96
@kimmonismus can now run something this capable inside Blender / Unreal on their own infra, keep unreleased worlds + asset libraries private, and still get near-frontier agent performance.
That combination of open weights + strong 3D/tool use + cheap inference feels way more important than the benchmark win by itself.
That combination of open weights + strong 3D/tool use + cheap inference feels way more important than the benchmark win by itself.
0
0
7
1,615
@kimmonismus Kimi really just pulled off something big without making a fuss about it. 😂
0
0
2
1,490
@kimmonismus Why are you so surprised, chubby? This was inevitable.
0
0
1
267
@kimmonismus Usage is insane, they don’t have the compute needed.
0
0
2
1,182
@kimmonismus what an amazing day for FREEDOM! open source DOMINATION!
0
0
0
175
@kimmonismus Being half the cost doesn't necessarily mean the model will actually be half the cost. If it outputs twice as many tokens as Opus 5.5 for the same task, then it costs exactly the same. I'll hold judgment until I actually get to try the model.
1
0
2
575
@kimmonismus Incredible value for the price, we need Chinese models to provide affordable alternatives.
0
0
0
281
@kimmonismus Very good, but its a shame that kimi gives the worst value subscription out of all of them.
0
0
0
1,065
@kimmonismus They should ban people in the US to prevent distillation since a US company called Anthropic is notorious for distilling the world.
0
0
0
607
@kimmonismus Where it beats it? In trusm me bro benchmarks? Have you tried it yourself? Where are the test runs? Wtf is this useless post?
0
0
0
148
@kimmonismus But who cares about API pricing? Hardly anyone uses the API anyway…
1
0
0
583
@kimmonismus They're distilling. still after the CoT, removal and ecrypted.. idk what they're going to be the excuse now rather than accep chinese labs are talented as well.
0
0
0
182
@kimmonismus weights aren't out yet, so i'd hold the local running talk until 4 Nov
0
0
0
34
@kimmonismus 50% cheaper than opus. still over my budget
0
0
7
1,782
@kimmonismus Worst then sonnet for the same price
0
0
0
125
@kimmonismus Im so upset at not joining Kimi when I had the chance
0
0
0
207
@kimmonismus Is this available already? Super hyped to try it out
0
0
1
1,554
@kimmonismus sounds great, but - moonshot's own evals tho, i'd wait for independent numbers. 🤠
We need an independent organisation!
We need an independent organisation!
0
0
0
302
