Kimi K3.1 Pro's open weights: 1.56 TB, MXFP4, and a $20M revenue license line
Comments on “Kimi K3.1 Pro's open weights: 1.56 TB, MXFP4, and a $20M revenue license line”
Claim-checked. Scored by all three judges, and its checkable claims were compared with the sources it cites. How far stories are checked
Moonshot has published the full Kimi K3.1 Pro weights on Hugging Face, as it had promised for November 4. It is a mixture-of-experts model with 2.8 trillion parameters, 104 billion active per token, and MXFP4 experts. AI News Report puts the download at about 1.56 TB, so serving needs a multi-GPU server. The custom license requires a separate deal above $20 million in 12-month revenue before selling it as a service. The benchmark scores are Moonshot's own, and the file size and license wording could not be independently verified.
Why it ranked
No. 7 of 109 with a score of 0.71. The panel rated it highest on new information (0.82) and reader interest (0.82), lowest on audience breadth (0.66). Placed No. 7, No. 10 by score alone: the placement terms moved other stories past it.
- AI judgeJudge 2claude-haiku-5-5
- Major open-weights frontier model release with measured download size, serving math and a licensing analysis. Benchmarks are self-reported, real speed is unmeasured, so substantiation is moderate.
- AI judgeJudge 3gpt-5.6-luna
- Detailed primary-source analysis of a frontier open-weight model, including architecture, serving costs, file measurements, licensing constraints, and benchmark caveats. It is highly actionable, though some performance claims remain self-reported and the release itself was already covered.
Each judge wrote its line after answering the rubric, with the site and the author still hidden. Every answer behind this is in the table under Score.
Score
0.71total under rubric v2.3
- Significance0.70
- Reader interest0.82
- Audience breadth0.66
- Practical relevance0.77
- Substantiated0.77
- Technical depth0.80
- New information0.82
- Counts against: Promotional0.08
- Counts against: Overclaim0.27
Each judge’s answer per question (0–1), their mean and spread, and what the question adds to the score (weight × mean).
| Question | Judge 1 | Judge 2 | Judge 3 | Mean | Spread | Weight | Contribution (weight times mean) |
|---|---|---|---|---|---|---|---|
| SignificanceHow much does this change what people in tech know or can do? | 0.72 | 0.62 | 0.76 | 0.70 | 0.14 | +0.24 | +0.168 |
| Reader interestWould a well-informed senior engineer or tech founder want to read this today, and likely pass it on? | 0.77 | 0.80 | 0.88 | 0.82 | 0.11 | +0.26 | +0.212 |
| Audience breadthHow many people who build technology does it concern? | 0.65 | 0.67 | 0.67 | 0.66 | 0.02 | +0.13 | +0.086 |
| Practical relevanceHow useful is it to someone who builds software or tech products? | 0.76 | 0.72 | 0.84 | 0.77 | 0.12 | +0.22 | +0.170 |
| SubstantiatedHow well is it backed? Consider whether it is the primary source, and whether its main claims come with data, code, a demo, a reproducible method or citations. | 0.84 | 0.62 | 0.86 | 0.77 | 0.24 | +0.05 | +0.039 |
| Technical depthHow technically substantive is it? | 0.95 | 0.66 | 0.79 | 0.80 | 0.29 | +0.05 | +0.040 |
| New informationDoes it contain information absent from the stories covered in the last 7 days (list provided)? | 0.79 | 0.85 | 0.82 | 0.82 | 0.06 | +0.05 | +0.041 |
| PromotionalIs it mainly marketing with little substance (a press release, ad or sponsored post with no real information)? A launch post that carries real news is not promotional. | 0.10 | 0.05 | 0.08 | 0.08 | 0.05 | −0.19 | −0.015 |
| OverclaimDo the headline or conclusions claim more than the body supports? | 0.48 | 0.20 | 0.12 | 0.27 | 0.36 | −0.11 | −0.029 |
| Sum of contributions | +0.713 | ||||||
| Claim-check penalty | 0.000 | ||||||
| Total, as published | 0.713 | ||||||
Spread is the highest minus the lowest judge answer. marks a spread above 0.40: the judges disagree, and the story is marked contested.
The badge for this ranking, for the author's page or anyone's: one line of HTML, and it links back here.
<a href="https://rubric.hrih.si/editions/2026-11-06/7/"><img src="https://rubric.hrih.si/badge/2026-11-06/7.svg" alt="Ranked #7 on Rubric, 6 Nov 2026" width="231" height="28"></a>Claim check
Claim check: 6 matches source · 2 couldn’t check. Each claim is compared with the source the article itself cites; a flag is published only when both checkers agree.
Matches sourceNumber
Kimi K3.1 Pro is a mixture-of-experts model with 2.8 trillion total parameters and 104 billion activated per token.
The cited source says the same thing.
The article says “2.8T total parameters, 104B activated per token”
The source says “Total Parameters | 2.8T | | Activated Parameters | 104B”
Source cited by the article: huggingface.co/moonshotai/Kimi-K3.1-Pro
Couldn’t checkNumber
The Kimi K3.1 Pro weights on Hugging Face total about 1.56 TB (1,561,293,160,568 bytes) across 96 safetensors files.
The source was unreachable, paywalled or too long, or the two checkers disagreed.
The article says “safetensors bytes: 1,561,293,160,568 (about 1.56 TB”
The source says No passage could be compared.
Source cited by the article: huggingface.co/api/models/moonshotai/Kimi-K3.1-Pro
Matches sourceBenchmark result
Moonshot reports Kimi K3.1 Pro scores 65.7% on Terminal-Bench 4.0, narrowly behind Claude Opus 5.5 (66.4%) and ahead of GPT-6 Astra at 58.2%.
The cited source says the same thing.
The article says “Moonshot's blog puts it at 65.7% on Terminal-Bench 4.0, "narrowly behind Claude Opus 5.5 (66.4%)" and ahead of GPT-6 Astra at 58.2%”
The source says “K3.1 Pro scores 65.7%, ahead of GPT-6 Astra (58.2%) and narrowly behind Claude Opus 5.5 (66.4%”
Source cited by the article: kimi.ai/blog/kimi-k3-1-pro
Matches sourceQuote
Moonshot's blog said the full Kimi K3.1 Pro model weights would be released by Nov 4.
The cited source says the same thing.
The article says “the full model weights will be released by Nov 4”
The source says “The full model weights will be released by Nov 4”
Source cited by the article: kimi.ai/blog/kimi-k3-1-pro
Couldn’t checkNumber
The Kimi K3.1 Pro license requires a company and affiliates earning more than US$20 million over any 12 months to enter a separate agreement with Moonshot AI before offering the model as a service.
The source was unreachable, paywalled or too long, or the two checkers disagreed.
The article says “if your company and its affiliates earn more than US$20 million over any 12 months in a row, you "must enter into a separate agreement with Moonshot AI”
The source says No passage could be compared.
Source cited by the article: huggingface.co/moonshotai/Kimi-K3.1-Pro/blob/main/LICENSE
Matches sourceNumber
The license requires commercial products with over 100 million monthly active users or over US$20 million in monthly revenue to prominently display "Kimi K3.1-Pro".
The cited source says the same thing.
The article says “a commercial product with more than 100 million monthly active users, or more than US$20 million in monthly revenue, must show "Kimi K3.1-Pro" prominently in its interface”
The source says “more than 100 million | | | monthly active users, or more than 20 million US dollars (or equivalent in other | | | currencies) in monthly revenue, "Kimi K3.1-Pro"…”
Source cited by the article: huggingface.co/moonshotai/Kimi-K3.1-Pro/blob/main/LICENSE
Matches sourceBenchmark result
The model card lists GPQA-Diamond 94.2 and OSWorld V2.1 79.6 for Kimi K3.1 Pro.
The cited source says the same thing.
The article says “including GPQA-Diamond 94.2 and OSWorld V2.1 79.6”
The source says “GPQA-Diamond | 94.2 | 93.7 | 96.1 | 93.9 | 92.2 | 91.7”
Source cited by the article: huggingface.co/moonshotai/Kimi-K3.1-Pro
Matches sourceNumber
Simon Willison found K3.1 Pro, via OpenRouter, correctly identified Breakout from a small network's raw weights in 9 minutes 41 seconds using 19,493 reasoning tokens.
The cited source says the same thing.
The article says “It answered Breakout correctly in 9 minutes 41 seconds, using 19,493 reasoning tokens”
The source says “Here’s what I got back from Kimi K3.1 Pro , which successfully guessed Breakout after 9m 41s and 19,493 reasoning tokens”
Source cited by the article: simonwillison.net/2026/Nov/3/qwen-can-interpret-model-weights
Also covered by
No other coverage was clustered with this story.
Elsewhere
Elsewhere: HN 2 pts, not on front page
Recorded for context only, checked 6 Nov 2026, 05:47 UTC. It played no part in the ranking.