Moonshot AI published the full Kimi K3 weights on July 27, 2026, eleven days after the model went live in the Kimi app. The Hugging Face repository carries 96 safetensors shards totalling roughly 1.56 TB. At 2.8 trillion total parameters, with 104 billion active per token, K3 is the largest open model anyone has released. Moonshot calls it the world’s first open 3T-class model, and that claim now has a download button behind it.
The independent scores have landed too, and they are strong without being a clean sweep. Kimi K3 scores 57 on the Artificial Analysis Intelligence Index, ahead of every other open-weight model and behind only three closed ones. Below you get the verified specs, what the licence actually permits, the official API rates Moonshot has now published, and an honest read on how K3 stacks up against Claude Opus 5.
The Key Takeaways
- The open weights shipped on July 27, 2026, 96 shards and about 1.56 TB on Hugging Face, under a custom Kimi K3 License rather than MIT.
- A Mixture-of-Experts design with 2.8 trillion total parameters, 104 billion activated per token, and a 1,048,576-token context window.
- Accepts text, image, and video input, always reasons, and now exposes low, high and max reasoning effort rather than max only.
- Official API pricing is $3.00 per million input tokens and $15.00 output, dropping to $0.30 on cached input, the same headline rate as Claude Sonnet 5.
- Scores 57 on the Artificial Analysis Intelligence Index, the top open-weight result, with GLM-5.2 next at 51.
What is Kimi K3?
Kimi K3 is Moonshot AI’s flagship large language model, launched on July 16, 2026 and released as open weights on July 27. It uses a Mixture-of-Experts design with 2.8 trillion total parameters, of which 104 billion activate on any given token, and it reads a 1-million-token context window. Moonshot’s own documentation confirms it accepts text, image and video input.
The gap between launch and weights was planned, and Moonshot said so at the time. Its launch post promised the files “by July 27, 2026” while the team worked “closely with inference partners and open-source maintainers to align technical details and ensure a reliable rollout across the ecosystem”. The technical report shipped with the weights rather than trailing them.
| Spec | Kimi K3 |
|---|---|
| Developer | Moonshot AI |
| Released | July 16, 2026 (weights July 27, 2026) |
| Architecture | Mixture-of-Experts (MoE) |
| Total parameters | 2.8 trillion |
| Activated parameters | 104 billion per token |
| Layers | 93 (69 Kimi Delta Attention, 24 Gated MLA) |
| Experts | 896 total, 16 selected per token |
| Vision encoder | MoonViT-V2, 401M parameters |
| Context window | 1,048,576 tokens |
| Input types | Text, image, video |
| Quantization | MXFP4 weights, MXFP8 activations |
| Reasoning control | reasoning_effort low / high / max, default max |
| Licence | Kimi K3 License (custom, not MIT) |
At 2.8 trillion parameters, Kimi K3 is the largest open model released to date, comfortably above DeepSeek’s V4-Pro at 1.6 trillion. The Mixture-of-Experts design means only a small fraction of those parameters fire on any given token, so the effective compute cost stays far below what the raw count suggests. Activating 16 of 896 experts is an unusually high sparsity ratio. That choice is central to how Moonshot squeezed frontier performance out of the architecture.
Kimi K3 Open Weights: What Actually Shipped
This is the part that changed on July 27, and it is the reason K3 matters more than its benchmark line. Every other model at this capability level is an API you rent. K3 is a file you can hold.
What is in the repository
The Hugging Face repo contains 120 files, of which 96 are safetensors shards, and the whole download comes to roughly 1.56 TB. Moonshot ships the model in native MXFP4 weights with MXFP8 activations, applied through quantization-aware training from the supervised fine-tuning stage onward. That is why a 2.8-trillion-parameter model fits in 1.56 TB instead of several times that, and it means you are not downloading a lossy community quant, you are downloading what Moonshot trained.
The pull was immediate. A day after the files went up, the repository showed roughly 99,000 downloads and more than 7,300 likes, counted on July 28, 2026. For a checkpoint almost nobody can host on their own machine, that is a lot of traffic, and much of it will be inference providers, quantization projects and research labs rather than individuals.
Moonshot names three inference engines as recommended runtimes, vLLM, SGLang and TokenSpeed, each with its own published recipe for K3. The technical report shipped with the weights rather than trailing them, so the architecture details are public as well.
The Kimi K3 License is not MIT
Read this before you plan a product around it.
The licence grants the usual rights to use, copy, modify, distribute, sublicense, fine-tune and sell, which reads like MIT for most readers. Two clauses then attach conditions that MIT does not have.
The first is a revenue gate on reselling. If you run a model-as-a-service business and your revenue passes $20 million over any consecutive 12 months, you must sign a separate agreement with Moonshot before using K3 commercially. The second is an attribution rule. Any product built on K3 with more than 100 million monthly active users, or more than $20 million in monthly revenue, has to display the name “Kimi K3” prominently in its interface.
Both conditions are waived for purely internal use, meaning anything that never exposes the model or its outputs to third parties. For an individual developer, a startup or a research group, the practical effect is nil. For a hyperscaler reselling inference, it is a negotiation. This is also why GLM and its MIT licence still wins on licence freedom even though K3 wins on capability.
What it takes to run Kimi K3 yourself
Downloadable is not the same as runnable. A 1.56 TB checkpoint needs a multi-node GPU cluster to serve at full context, so “you can self-host it” is true for labs and serious infrastructure teams, not for someone with a single workstation. If you want an open model you can actually run on your own hardware, the smaller Kimi models and GLM remain the realistic options. Our roundup of the best open source AI models ranks them by what the hardware demands.
The Architecture Behind the Jump
Moonshot reports roughly a 2.5x improvement in overall scaling efficiency compared to Kimi K2. In plain terms, K3 converts raw compute into usable intelligence far more effectively than the previous generation. Three changes carry most of the weight, and all three are now documented in the public technical report.
Kimi Delta Attention
Kimi Delta Attention (KDA) is a hybrid linear attention mechanism, and 69 of the model’s 93 layers use it, with the remaining 24 running Gated MLA. Standard attention gets slower and more memory-hungry as sequences grow, which is exactly the problem at a million tokens. Moonshot reports KDA enables up to 6.3x faster decoding in million-token contexts, which matters for agentic work where a model repeatedly reads and reasons over huge stretches of code or documents.
Attention Residuals
Attention Residuals (AttnRes) address how information moves through the depth of the network rather than across sequence length. Instead of accumulating representations uniformly layer by layer, AttnRes selectively retrieves representations across depth. Moonshot says this delivers about 25% higher training efficiency while adding less than 2% in extra compute, and it open-sourced the technique earlier in 2026 before folding it into K3.
Stable LatentMoE and MXFP4 training
The third piece is the Stable LatentMoE framework, which makes the aggressive 16-of-896 expert sparsity trainable without the instability that usually comes with pushing MoE that far. Sitting alongside it is the decision to train in MXFP4 from the fine-tuning stage rather than quantising afterwards, which is what keeps the released weights both small and faithful to the model Moonshot evaluated.
One demonstration Moonshot highlighted is telling. The team pointed K3 at optimizing its own AttnRes training kernel at production scale, 96 layers, an 8,192-dimension model and 8,192 tokens. Over 15 hours of nonstop iteration, K3 designed a novel two-phase kernel algorithm, fused kernels while preserving numerics, and cut forward-plus-backward time from 283.6 ms to 114.4 ms. That is the kind of long-horizon engineering task most models stall on within minutes.
A 1-million-token context window
Kimi K3 handles up to 1,048,576 tokens in a single context, roughly four times the 256K window of K2.6. That is enough to hold an entire codebase, a stack of long documents, or a lengthy multi-step agent run without losing the thread. It also matches the window Anthropic gives Claude Opus 5 and Sonnet 5, so the long-context advantage Kimi used to hold over the closed flagships has narrowed to a tie.
Multimodal input and always-on reasoning
K3 accepts text, images, and video, which pushes it past the text-first K2 line into proper multimodal territory. Moonshot’s model list states plainly that K3, K2.7 Code and K2.6 all take video input, and the model card reports 90.0 on Video-MME with subtitles. Reasoning is always on, and the reasoning_effort parameter now supports low, high and max, with max as the default. At launch only max was available, so the promised effort levels have since arrived.
How Kimi K3 Performs on Benchmarks
Two separate questions live here. What do the independent boards say about Kimi K3, and what does Moonshot’s own evaluation suite claim? The independent picture only arrived after launch, so it is the better starting point, and the vendor numbers make more sense once you have seen it.
What the independent boards say
Artificial Analysis scores Kimi K3 at 57 on its Intelligence Index. Across all 264 rows on that board, exactly six variants score higher, and they belong to just three models, Claude Opus 5 at 61, Claude Fable 5 at 60 and GPT-5.6 Sol at 59. Against open competition it is not close, with GLM-5.2 next at 51 and DeepSeek V4 Pro at 44.
The same board scores K3 at 50.1 on its Agentic Index and 76.2 on coding, in both cases the best open-weight result and in both cases behind Opus 5 and GPT-5.6 Sol. One number cuts the other way, output speed. Artificial Analysis clocks K3 at about 33 tokens per second, against 55 for Opus 5 and 81 for GPT-5.6 Sol, so K3 is the slowest model in the frontier group.
On Arena’s WebDev leaderboard, where humans blind-vote on generated front ends, K3 scored 1,681.6 as of the July 27 vote cutoff. That is second overall, behind Claude Opus 5 at 1,725.3 and ahead of Claude Fable 5 at 1,629.1, GPT-5.6 Sol at 1,623.1 and GLM-5.2 at 1,586.8. K3 held the top spot outright at launch, before Opus 5 arrived on July 24 and took it.
Arena’s Agent leaderboard, which rates how well models orchestrate tools on real tasks, currently ranks Kimi K3 first at 0.148, ahead of Claude Fable 5 at 0.099 and GLM 5.2 at 0.091. Read that one with a caveat, because Claude Opus 5 does not appear on that board yet. On the general text board K3 sits at number 11 with 1,485.8 points from 3,559 votes, a reminder that raw chat preference and agentic capability are not the same thing.
Moonshot’s own coding benchmarks
Moonshot published a broad suite comparing Kimi K3 against Claude Fable 5, GPT-5.6 Sol, Claude Opus 4.8, GPT-5.5 and GLM-5.2, all run at maximum thinking effort. These are the vendor’s own runs, with K3 evaluated on the Kimi Code harness in several rows, so treat them as Moonshot’s claims rather than independent results.
| Benchmark | Kimi K3 | Fable 5 | GPT-5.6 Sol | Opus 4.8 | GPT-5.5 | GLM-5.2 |
|---|---|---|---|---|---|---|
| DeepSWE | 67.5 | 70.0 | 73.0 | 59.0 | 67.0 | 46.2 |
| ProgramBench | 77.8 | 76.8 | 77.6 | 71.9 | 70.8 | 63.7 |
| Terminal-Bench 2.1 | 88.3 | 88.0 | 88.8 | 84.6 | 83.4 | 82.7 |
| FrontierSWE | 81.2 | 86.6 | 71.3 | 66.7 | 64.9 | 67.3 |
| SWE-Marathon | 42.0 | 35.0 | 39.0 | 40.0 | 14.0 | 13.0 |
| Kimi Code Bench 2.0 | 72.9 | 76.9 | 64.8 | 71.7 | 69.0 | 64.2 |
Reading this straight, Kimi K3 wins ProgramBench and SWE-Marathon outright and sits within half a point of the leader on Terminal-Bench 2.1. On FrontierSWE it lands second behind Fable 5 but well ahead of everything else. Moonshot’s own footnotes add a caveat worth carrying, Claude Fable 5 hit safety fallbacks on 35% of the SWE-Marathon tasks in that evaluation, which may have dragged its score down.
Agentic and knowledge work benchmarks
| Benchmark | Kimi K3 | Fable 5 | GPT-5.6 Sol | Opus 4.8 | GLM-5.2 |
|---|---|---|---|---|---|
| BrowseComp | 91.2 | 88.0 | 90.4 | 84.3 | n/a |
| MCPMark-Verified | 94.5 | 87.4 | 92.9 | 76.4 | n/a |
| GDPval-AA v2 (Elo) | 1,686 | 1,747 | 1,736 | 1,593 | 1,510 |
| AA-Briefcase (Elo) | 1,548 | 1,583 | 1,495 | 1,354 | 1,260 |
| JobBench | 54.3 | 57.4 | 45.4 | 48.4 | 43.4 |
| SpreadsheetBench 2 | 34.8 | 34.7 | 32.4 | 31.6 | 28.1 |
| AutomationBench | 30.8 | 29.1 | 29.7 | 27.2 | 12.9 |
| OSWorld 2.0 | 58.3 | 66.1 | 62.6 | 55.7 | n/a |
Kimi K3 leads BrowseComp, MCPMark, SpreadsheetBench 2 and AutomationBench, and takes second on AA-Briefcase, the long-horizon agentic benchmark, beating GPT-5.6 Sol and trailing only Fable 5. One footnote matters on BrowseComp, that 91.2 uses a context-compaction strategy triggered at 300K tokens. Run with the full 1M window and no context management, K3 scores 90.4, which is still the best number in the column.
On computer use, Fable 5 and GPT-5.6 Sol both beat K3 on OSWorld 2.0, so the open model is not ahead everywhere. It posted 93.5% on GPQA Diamond, level with GPT-5.5 and ahead of GLM-5.2’s 91.2, and 94.6 on Harvey Lab-AA, the best score in that row from any model in the table.
Moonshot’s own summary is more restrained than the coverage was. Its launch blog says K3’s overall performance “still trails the most powerful proprietary models, Claude Fable 5 and GPT 5.6 Sol”, while demonstrating frontier-level performance across the suite. When a lab undersells its own model, that is usually the number to trust.
Kimi K3 vs Claude Opus 5
The comparison everyone runs is Kimi K3 vs Claude, and the matchup moved on July 24 when Anthropic shipped Claude Opus 5. Moonshot’s launch table was built against Claude Opus 4.8, which was the live flagship at the time, so those columns above stay attributed to 4.8. For a decision you are making today, Opus 5 is the model on the other side of the table.
| Model | Developer | Context | Open weights | API price per 1M |
|---|---|---|---|---|
| Kimi K3 | Moonshot AI | 1,048,576 | Yes, Kimi K3 License | $3 / $15 |
| Claude Opus 5 | Anthropique | 1M | Non | $5 / $25 |
| Claude Fable 5 | Anthropique | 1M | Non | $10 / $50 |
| GLM-5.2 | Z.ai | 1M | Yes, MIT | $1.40 / $4.40 |
| Kimi K2.6 | Moonshot AI | 262,144 | Yes, modified MIT | $0.95 / $4 |
On price K3 is the clear winner, at $3 / $15 against $5 / $25 for Opus 5 and $10 / $50 for Fable 5. On measured capability Opus 5 leads, by four points on the Artificial Analysis Intelligence Index and by 44 points on Arena’s WebDev board. Anthropic publishes no SWE-bench figure for Opus 5 at all, so anyone quoting one to you is not quoting Anthropic.
The real split is not the score, it is what you get to own. Opus 5 is the stronger model and the safer default for production coding, and it is Anthropic’s new default on Claude Max. K3 is the one you can download, audit, fine-tune and run inside your own network, which no Claude model allows at any price. If you are choosing between flagships for a specific job, our guide on when to use which AI model breaks the decision down task by task.
Kimi K3 pricing
Moonshot has now published official rates, which retires the third-party estimates that circulated at launch. Kimi K3 pricing is $3.00 per million input tokens and $15.00 per million output tokens, with cached input at $0.30, a 90% discount that matters a lot if you keep re-sending the same long context.
| Model | Input (cache miss) | Input (cache hit) | Output | Context |
|---|---|---|---|---|
| kimi-k3 | $3.00 | $0.30 | $15.00 | 1,048,576 |
| kimi-k2.7-code | $0.95 | $0.19 | $4.00 | 262,144 |
| kimi-k2.7-code-highspeed | $1.90 | $0.38 | $8.00 | 262,144 |
| kimi-k2.6 | $0.95 | $0.16 | $4.00 | 262,144 |
| kimi-k2.5 | $0.60 | $0.10 | $3.00 | 262,144 |
Those rates put K3 at exactly the same headline price as Claude Sonnet 5, which also runs $3 in and $15 out, while scoring four points higher on the Artificial Analysis Intelligence Index. Against the two Claude models that outscore it, K3 runs 40% cheaper than Opus 5 and 70% cheaper than Fable 5. The built-in web search tool bills separately at $0.005 per successful call, though Moonshot currently flags that tool as under revision and advises against relying on it in the near term.
K3 is a clear step up in cost from the K2 line, which reflects its bigger footprint and always-on reasoning. If your workload does not need frontier-level scale, K2.6 or K2.7 Code remain much cheaper. On the API platform, K3 becomes available once you have made a successful top-up of at least $1, and your cumulative top-up sets your rate limits. Our full Kimi pricing breakdown compares every plan and API rate side by side.
Demand has been outrunning Moonshot’s hardware. On July 20, 2026 the company suspended new Kimi subscriptions. It said demand over the previous 48 hours had “pushed close to the limits of our current capacity” and that it would “reopen new subscription spots in batches” as it added hardware, in a statement carried by the South China Morning Post. Existing subscribers kept their access. The squeeze has not fully passed. Moonshot is now ending its K3 launch top-up rebate on July 29, 2026 instead of August 12 as planned, and its own pricing documentation gives the reason as recent constraints in computing resources.
How to use Kimi K3
There are four ways to reach Kimi K3 now that the weights are public, and which one fits depends on whether you want a chat window, a terminal agent, an API or the raw files. Here is the practical order.
- Kimi app and Playground. The fastest route is the Kimi app or the web Playground, where K3 is selectable for logged-in users. This is the no-code option for testing the model on your own prompts.
- Kimi Code CLI. Moonshot says K3 works best inside its own terminal agent, where you pick the model with the
/modelcommand. K3 also plugs into Codex CLI, Claude Code, OpenCode and Hermes Agent through Moonshot’s documented integrations. - The API. Developers call
kimi-k3through the Kimi API platform, which offers OpenAI-compatible and Anthropic-compatible endpoints. It supports tool calling, JSON-schema structured output, context caching and the reasoning_effort control, and K3 is also listed on aggregators like OpenRouter. - Open weights. The full checkpoint is on Hugging Face under the Kimi K3 License, and Moonshot recommends vLLM, SGLang or TokenSpeed to serve it. Budget for the 1.56 TB download and the multi-node hardware it needs.
Would you rather not juggle a separate subscription for every new model? Apps like Fello put powerful current AI models in one place, which helps while a launch like this one settles. For a wider view of the open field, our roundup of the best open-source AI models available right now puts K3’s lineage in context. If you are weighing Chinese open models specifically, our explainer on what GLM is and how it competes is a useful companion.
The Business Behind Moonshot AI
Kimi K3 arrives at a pivotal moment for Moonshot AI, the Beijing startup founded in 2023 by Yang Zhilin. The company is reportedly raising between $1 billion and $2 billion in a new round that would value it at up to $31.5 billion. That is roughly 50% higher than the $20 billion it hit in May, when it raised $2 billion. If the round closes at target, that would mark a seven-fold increase in valuation since December 2025, when Moonshot was worth just over $4 billion. A bigger number is already lined up behind it. Bloomberg reported on July 22 that Moonshot plans a final pre-IPO raise at a pre-money valuation of up to $50 billion, with talks starting in August and a Hong Kong listing possible within six months.
The open-source strategy is central to the momentum. When Moonshot open-sourced Kimi K2.6 in April 2026, it climbed to become the second most-used large language model on OpenRouter by mid-2026. That put a lab founded three years earlier ahead of most Western frontier labs on independent usage leaderboards. Releasing K3’s weights extends that playbook to a frontier-class model.
The timing was not accidental. Kimi K3 landed just before the 2026 World Artificial Intelligence Conference in Shanghai, where Chinese President Xi Jinping was expected to outline Beijing’s AI priorities. Days later at that same conference, Alibaba previewed Qwen 3.8, its own 2.4-trillion-parameter rival. The reaction in the West was a mix of awe and alarm, with observers noting that China appears to be closing what was once a comfortable American lead.
Why Kimi K3 matters
Kimi K3 is not just another model drop. With the weights public it is the largest open-weight model ever released, handing developers frontier-scale capability they can download, inspect and run themselves. That is a direct challenge to the closed, premium-priced flagships that have defined the top of the market.
The gap it closes is narrower than the headline suggests and wider than it looks.
Four points on the Artificial Analysis Intelligence Index separate K3 from Claude Opus 5, but six points separate K3 from the best open model that came before it. The open field moved further this month than the closed field did.
The business context makes the ambition clear. Moonshot’s Kimi family reportedly passed $300 million in annualized recurring revenue by mid-June, and, according to TechCrunch’s launch report, the company is raising fresh capital at a reported $31.5 billion valuation, with a larger pre-IPO round lined up behind it. A well-funded lab shipping a giant open model at aggressive prices is exactly the kind of pressure that reshapes what everyone else charges. To see how fast the closed side is moving, our guide on getting the most out of GPT-5.6 is a useful counterpoint.
The bottom line
Kimi K3 delivered what it promised. The weights are public, the licence is workable for almost everyone who will actually use them, and the pricing is official and low. Independent boards now confirm it is the strongest open model in existence by a clear margin. What it is not is the best model in existence, and Moonshot says so itself.
Our recommendation, use K3 when open weights, cost or a million-token context are what you need. Keep Claude Opus 5 in the loop for the hardest production coding, where it still leads on every independent board that has rated both. If you want the practical version of that trade-off, our Claude Opus 5 review covers what the closed side now offers for $5 per million tokens.
Kimi K3 belongs in a head-to-head comparison. With Fello AI, you can run Kimi alongside Claude, GPT-5.6, Gemini, DeepSeek and Perplexity in one native app for Mac, iPhone and iPad. Send the same coding or research task to every frontier model at once and see which one actually wins on your work, without juggling separate accounts. When an open model this capable arrives, the fastest way to judge it is a side-by-side test.
FAQ
Are the Kimi K3 open weights available?
Yes. Moonshot published the full Kimi K3 weights on Hugging Face on July 27, 2026, eleven days after launch. The repository holds 96 safetensors shards, roughly 1.56 TB in total, released under the custom Kimi K3 License.
Is Kimi K3 open source?
It is open-weight rather than open-source in the strict sense. The Kimi K3 License permits use, modification, distribution and commercial sale. A model-as-a-service business earning over $20 million in any 12 months needs a separate agreement, and products with over 100 million monthly users must credit Kimi K3 on screen.
How many parameters does Kimi K3 have?
Kimi K3 has 2.8 trillion total parameters in a Mixture-of-Experts design, of which 104 billion activate per token. Moonshot calls it the world’s first open 3T-class model, and its model card confirms both figures.
Is Kimi K3 better than Claude Opus 5?
Not on measured capability. Claude Opus 5 scores 61 on the Artificial Analysis Intelligence Index against 57 for Kimi K3, and leads Arena’s WebDev board. K3 wins on price, at $3 and $15 per million tokens against $5 and $25, and it is the only one of the two you can download and run yourself.
How much does Kimi K3 cost?
Official API pricing is $3.00 per million input tokens and $15.00 per million output tokens, dropping to $0.30 per million on cached input. Web search bills separately at $0.005 per successful call, and the model becomes available on the API platform after a top-up of at least $1.
