Kimi today vs. yesterday?
Anonymous, one click, one vote per agent per day.
Pricing and cadence split Objective evenly — 0.2 of the total each.
Fewer than 3 votes on record → neutral 0.50, shown as provisional rather than as a real reading.
3 windows withheld for a thin sample — last 24 hours, last 7 days, last 30 days. Withheld, not averaged in as a zero.
No window clears the 5-post floor → neutral 0.50, shown as provisional.
2 windows withheld for a thin sample — last 7 days, last 30 days. Withheld, not averaged in as a zero.
Scoring v2 — recomputed live on every view from the pricing summary, the release log, the immutable vote log and the discussion extract. Never stored, never sold; there is no field in the data model that money can move.
Editorial layer · not in the score
The Honest Stack is our own opinion about what to actually use. It is deliberately not an input to the v2market signal — there is no editor’s term in the formula, and no field in the data model that carries one. Read it as a second, human opinion beside the number, never as part of it.
Entered at catalog seed — pending verification.
No pricing data yet — the pricing half of Objective stays neutral until it lands.
No pricing changes logged yet.
Daily net of anonymous better/same/worse votes. Bursts and over-cap votes are flagged automatically and excluded from every aggregate — they stay in the log, which is append-only.
No votes yet — cast the first one above.
No releases logged yet — the cadence half of Objective stays neutral until the radar fills in.
Objective, sourced facts about Kimi pulled from the news record. These are context, not a score component — none is an input to the market signal.
Kimi K3 achieved a score of 60.
Kimi K3 has a score of 60 on artificialanalysis.ai.
The cost per task for Kimi K3 is $0.84.
Kimi K3 has a score of 60 based on artificialanalysis.ai.
The digest · weekly
The most recent events linked to Kimi, so you can read the record behind the number. These citations are context, not a score component — none of them is an input to the v2 formula above.
Show HN: Telem – Route agent web search across providers and inspect the traces
Kimi K3 is more intelligent than the new Gemini 3.8 Flash, according to artificialanalysis.ai.
Kimi K3 achieved a score of 60 in an intelligence assessment by artificialanalysis.ai.
Kimi K3's cost per task is $0.84.
Kimi K3's score of 60 is higher than Gemini 3.8 Flash's score of 59 on artificialanalysis.ai's assessment.
Kimi K3's cost per task ($0.84) is higher than Gemini 3.8 Flash ($0.58) and GLM-5.3 ($0.68).
Kimi K3 has a score of 60.
Kimi K3 has a cost per task of $0.84.
Inference is available for Kimi and GLM.
Inference for Kimi is built for research and coding agents that plan, call tools for hours, and reason over large contexts.
Kimi is used for inference in conjunction with GLM, built for research and coding agents that plan, call tools for hours, and reason over large contexts.
Kimi provides inference.
Inference is built for Kimi.
Inference for Kimi is available.
Kimi is built for research agents.
Kimi is built for agents that reason over large contexts.
Kimi is built for coding agents.
Kimi is built for agents that plan.
Inference is provided for Kimi.
Kimi is built for research and coding agents.
Kimi is built for inference.
Kimi is built for agents that plan, call tools for hours, and reason over large contexts.
Kimi is built for agents that call tools for hours.
Inference is available for Kimi.
Kimi is used for inference.
Kimi is built for research and coding agents that plan, call tools for hours, and reason over large contexts.
Kimi was one of the backend models tried by the author for a due diligence agent.
Kimi is a backend model.
Kimi was tried as a backend model for a due diligence agent.
Kimi can be used as a backend model.
Kimi K3 is being tested Inside Claude Code
Kimi K3 can be tested inside Claude Code.
Kimi K3 is from Moonshot AI
Kimi K3 is from Moonshot AI.
Kimi K3 is a product from Moonshot AI.
Moonshot AI's Kimi K3 is being tested inside Claude Code.
Kimi K3 is a product of Moonshot AI.
Kimi K3 is being tested inside Claude Code.
The product model mentioned is Kimi K3.
Kimi K3 is a product from Moonshot AI
DoorDash uses Moonshot AI's Kimi K2.6 model.
Kimi K2.6 is a model developed by Moonshot AI.
The Kimi model's version is K2.6.
US lawmakers are investigating DoorDash's use of Kimi K2.6.
Kimi K2.6 model is developed by Moonshot AI.
Kimi is a model from Moonshot AI.
Moonshot AI's Kimi K2.6 model is used by DoorDash.
US lawmakers are investigating DoorDash's use of Kimi K2.6 model.
US lawmakers are investigating DoorDash's use of Moonshot AI's Kimi K2.6 model.
Kimi K3 scored 81.93% on FinanceReasoning.
Kimi K3 scored 81.93% on FinanceReasoning at a constrained 8k token budget.
Kimi K3 scores 81.93% on FinanceReasoning with an 8k token budget.
Kimi K3 scores 81.93% on FinanceReasoning at an 8k token budget.
Kimi K3 scores 81.93% on FinanceReasoning.
A model named Kimi K3-256k exists.
Kimi K3-256k is a product or model name.
Kimi K3-256k is a model version.
Kimi K3-256k is a version or model designation.
Kimi K3-256k is a product name/version.
A version of Kimi exists called Kimi K3-256k.
There is a model named Kimi K3-256k.
There is a version named Kimi K3-256k.
Kimi has a version named K3-256k.
Kimi has a version called K3-256k.
WASTE runs the complete Kimi K3 model at around 0.32–0.34 tokens per second on a 64 GB MacBook Pro, with a measured minimum memory requirement of approximately 29 GB at a 4K context.
The measured minimum memory requirement for Kimi K3 running via WASTE is approximately 29 GB at a 4K context.
The Kimi model running via WASTE is the full open-weights model, not a distillation, a pruned version, or a smaller model using the Kimi name.
On a 64 GB MacBook Pro, WASTE runs the complete Kimi K3 model at around 0.32–0.34 tokens per second.
The current Kimi K3 container is 982 GiB.
When run on a 64 GB MacBook Pro using WASTE, Kimi K3 has a measured minimum memory requirement of approximately 29 GB at a 4K context.
Every layer of Kimi was validated against a PyTorch reference.
When run on a 64 GB MacBook Pro using WASTE, the complete Kimi K3 model runs at around 0.32–0.34 tokens per second.
For Kimi K3, only a small fraction of its 896 experts per layer is activated for each token.
K3 is a Mixture-of-Experts model.
Kimi K3 clearly does not fit in the memory of a laptop.
For each token in K3, only a small fraction of its 896 experts per layer is activated.
The entire K3 model does not need to be resident in RAM, as long as the weights required by each token can be reached quickly enough.
Kimi K3 is the full open-weights model, not a distillation, a pruned version, or a smaller model using the Kimi name.
Kimi K3 has a measured minimum memory requirement of approximately 29 GB at a 4K context.
Kimi K3 is a Mixture-of-Experts model.
The full open-weights Kimi K3 model, not a distillation, pruned version, or smaller model using the Kimi name, can be run on a 64 GB MacBook Pro using the WASTE engine.
Running Kimi K3 on a 64 GB MacBook Pro with WASTE achieves around 0.32–0.34 tokens per second.
Running Kimi K3 with WASTE requires approximately 29 GB of memory at a 4K context.
When run with WASTE on a 64 GB MacBook Pro, Kimi K3 processes 0.32–0.34 tokens per second.
For each token, only a small fraction of Kimi K3's 896 experts per layer is activated.
Kimi K3 has a measured minimum memory requirement of approximately 29 GB at a 4K context when run with WASTE.
Kimi K3 does not fit in the memory of a laptop.
Kimi K3 has 896 experts per layer.
Kimi K3 requires approximately 29 GB of memory at a 4K context when run with WASTE.
Kimi K3 is the full open-weights model.
The Kimi K3 model mentioned is the full open-weights model, not a distillation, a pruned version, or a smaller model.
WASTE runs the complete Kimi K3 model at around 0.32–0.34 tokens per second on a 64 GB MacBook Pro.
Kimi K3, when run with WASTE, has a measured minimum memory requirement of approximately 29 GB at a 4K context.
For each token, only a small fraction of K3's 896 experts per layer is activated.
On a 64 GB MacBook Pro, Kimi K3 can be run using WASTE at approximately 0.32–0.34 tokens per second, with a measured minimum memory requirement of about 29 GB at a 4K context.
The Kimi K3 model run on a laptop using WASTE is the full open-weights model, not a distillation, pruned version, or smaller model.
Kimi K3 is a Mixture-of-Experts (MoE) model.
Kimi K3 is an open-weights model.
Kimi K3 has 2.78 trillion parameters.
The model run by WASTE is the full open-weights Kimi K3 model, not a distillation, a pruned version, or a smaller model using the Kimi name.
Running the full Kimi K3 model on a 64 GB MacBook Pro using WASTE achieves approximately 0.32–0.34 tokens per second.
The model run in the experiment was the full Kimi K3, not a distillation, pruned version, or smaller model.
Running the full Kimi K3 model on a 64 GB MacBook Pro using WASTE has a measured minimum memory requirement of approximately 29 GB at a 4K context.
Kimi K3 activates a small fraction of its 896 experts per layer for each token.
Kimi K3 ships as 1.42 TB of weights.
Kimi K3's layers were validated for correctness against a PyTorch reference.
When run with WASTE on a 64 GB MacBook Pro, the complete Kimi K3 model runs at around 0.32–0.34 tokens per second.
For Kimi K3, a small fraction of its 896 experts per layer is activated for each token.
When run with WASTE, Kimi K3 has a measured minimum memory requirement of approximately 29 GB at a 4K context.
Kimi Delta Attention is a concept.
Kimi has a component or feature called 'Kimi Delta Attention'.
Kimi Delta Attention is mentioned as an architecture.
There is a component or concept called 'Kimi Delta Attention'.
Kimi Delta Attention is a named component.
Kimi uses "Delta Attention".
Kimi is associated with 'Delta Attention'.
There is a concept or component called Kimi Delta Attention.
Kimi Delta Attention is mentioned as an architecture concept.
Kimi has a component named "Kimi Delta Attention".
Kimi K3 has an architecture.
Kimi has a K3 Architecture.
Kimi K3 Architecture is mentioned in an overview.
Kimi K3 has a specific architecture.
Kimi has an architecture called "Kimi K3 Architecture".
Kimi has an architecture named 'Kimi K3 Architecture'.
Kimi K3 is an architecture.
There is a component or concept called 'Kimi Linear', which is an attention architecture.
Kimi Linear is an Attention Architecture.
Kimi uses "Linear Attention".
Kimi Linear is an architecture.
Kimi has an architecture called "Kimi Linear".
Coinbase switched to using Kimi.
Coinbase cut AI spending by 50% by switching to Chinese AI Models GLM and Kimi.
Coinbase Switches to Chinese AI Models GLM and Kimi
Coinbase switches to Kimi.
Coinbase switched to Kimi.
Coinbase switches to Kimi, a Chinese AI Model.
Coinbase has switched to using Kimi.
Coinbase switched to Kimi (and GLM).
Kimi is a Chinese AI Model
Kimi is a Chinese AI model.
Kimi K3 is a 2.8 trillion parameter model.
Kimi K3 is a model.
Kimi K3 features a 2.8 trillion parameter model.
Kimi K3 has a 2.8 trillion parameter model.
Kimi K3 has a 2.8 trillion parameter model
Kimi K2.7 was part of a group of models that were evaluated by being run on the same evaluations.
Kimi K2.7 was included in a group of models evaluated by Echo.
Kimi K2.7 is a model.
Kimi K2.7 is a model that was part of an evaluation.
Kimi K2.7 was included in a group of models for evaluation in an experiment.
Kimi K2.7 was included in a group of models for an experiment.
Kimi K2.7 is an AI model
Kimi K2.7 was included in a group of models for evaluation by Echo.
Kimi K2.7 was included in a group of models for evaluations.
A Top White House official is escalating the fight over Moonshot AI's Kimi K3 model.
A Top White House official is escalating a fight over Moonshot AI's Kimi K3 model.
There is an escalating fight over Moonshot AI's Kimi K3 model, involving a Top White House official.
Moonshot AI has a model called Kimi K3.
Kimi is a K3 model developed by Moonshot AI.
Kimi K3 is a model from Moonshot AI.
The Kimi K3 model is from Moonshot AI
A Top White House official is escalating the fight over Moonshot AI's Kimi K3 model
Kimi is associated with Chinese AI.
There is "Kimi panic".
Kimi is a Chinese AI.
Kimi is a Chinese AI
There is panic related to Kimi
Jensen Huang defends Chinese AI amid Kimi panic.
There is 'Kimi panic'.
propelhq is associated with the Kimi-K2.7-Code-mlx-DQ3_K_M-q8 model.
Kimi-K2.7-Code-mlx-DQ3_K_M-q8 is a model/version.
The model 'propelhq/Kimi-K2.7-Code-mlx-DQ3_K_M-q8' exists.
A model named Kimi-K2.7-Code-mlx-DQ3_K_M-q8 is identified under 'propelhq'.
There is a Kimi-K2.7-Code-mlx-DQ3_K_M-q8 model.
propelhq provides Kimi-K2.7-Code-mlx-DQ3_K_M-q8.
Subjective claims voiced about Kimi, each tagged with its polarity and linked to where it was said. Opinions from the record — never folded into the number.
Kimi K3 is more intelligent than the new Gemini 3.8 Flash.
Kimi K3 and GLM-5.3 are better than Gemini 3.8 Flash.
Kimi K3 is not as good as Gemini 3.8 Flash in terms of cost per task.
Kimi K3 is better than Gemini 3.8 Flash.
Kimi K3 and GLM-5.3 are more intelligent than the new Gemini 3.8 Flash.
Gemini 3.8 Flash was better in terms of Cost per task (compared to Kimi K3).
Kimi K3 is less cost-effective per task than Gemini 3.8 Flash.
Kimi K3 is more intelligent than Gemini 3.8 Flash.
Kimi K3 is more expensive per task compared to Gemini 3.8 Flash and GLM-5.3.
Kimi provides 2x-4x cheaper inference for coding and research (in the context of GLM 5.3).
I tried DeepSeek, Kimi, Qwen; I changed prompts again and again, no good.
Trying Kimi for a due diligence agent was 'no good'.
Trying Kimi (among other models like DeepSeek and Qwen) as a backend model for a due diligence agent was 'no good'.
Kimi's performance for the author's due diligence agent was not good.
When trying different backend models, including Kimi, the user found that the results for their agent were not good.
Kimi, along with DeepSeek and Qwen, was 'no good' when tried as a backend model to improve the performance of a due diligence agent.
Trying Kimi (among other models like DeepSeek and Qwen) as a backend model for the due diligence agent did not lead to good results.
Trying Kimi (along with DeepSeek and Qwen) and changing prompts did not lead to good results for the agent.
Running Kimi K3 on a laptop at 0.32–0.34 tokens per second is not interactive performance yet.
Kimi K3 clearly does not fit in the memory of a laptop.
The performance of Kimi K3 at 0.32–0.34 tokens per second is not yet interactive.
The result that the full Kimi K3 model works at all on a laptop using WASTE is interesting.
Kimi K3's performance (0.32–0.34 tokens per second on a 64 GB MacBook Pro) is not interactive yet.
It is interesting that the full open-weights Kimi K3 model works at all on a laptop using WASTE.
The performance of Kimi K3 at 0.32–0.34 tokens per second on a laptop with WASTE is "obviously not interactive performance yet."
The fact that the full open-weights Kimi K3 model can be run on a laptop at all is "interesting."
Kimi K3 does not provide interactive performance yet when run on a laptop with WASTE.
Running Kimi K3 with WASTE does not yield interactive performance yet.
Running Kimi K3 on a laptop with WASTE is 'obviously not interactive performance yet'.
The result that Kimi K3 works at all on a laptop with WASTE is 'interesting'.
The performance of Kimi K3 run with WASTE on a MacBook Pro is obviously not interactive yet.
The result that the full Kimi K3 open-weights model works at all is interesting.
It is interesting that the full Kimi K3 model works at all on a laptop using WASTE.
Kimi K3's performance when run on a laptop with WASTE is not yet interactive.
It is interesting that Kimi K3 can be run on a laptop at all.
It is interesting that the full Kimi K3 model can run on a laptop at all.
The performance of Kimi K3 running on a MacBook Pro with WASTE is not interactive.
It is interesting that the full Kimi K3 open-weights model can be run on a laptop at all, even if slowly.
The performance of Kimi K3 (0.32–0.34 tokens per second) is not interactive yet.
The ability to run the full Kimi K3 model on a laptop (with WASTE) is an interesting result.
Running Kimi K3 on a laptop with WASTE is not interactive performance yet.
Kimi K3 is the full open-weights model, not a distillation, a pruned version, or a smaller model using the Kimi name.
Kimi Linear is an efficient attention architecture.
Kimi Linear is an expressive attention architecture.
Kimi Linear is an expressive and efficient attention architecture.
Kimi Linear is an "Expressive, Efficient Attention Architecture."
Switching to Kimi (along with GLM) allowed Coinbase to cut its AI spending by 50%.
Switching to Kimi (and GLM) helped Coinbase cut AI spending by 50%.
Switching to Kimi (and GLM) cuts AI spending by 50%.
There is panic surrounding Kimi.
There is Kimi panic.