Kimi K3.1 Pro model weights are out
Comments on “Kimi K3.1 Pro model weights are out”
Claim-checked. Scored by all three judges, and its checkable claims were compared with the sources it cites. How far stories are checked
Moonshot AI released the weights for Kimi K3.1 Pro, a 2.8T-parameter mixture-of-experts model with 104B active parameters, a 1M-token context and MXFP4 quantization-aware training. It swaps K3's full-attention layers for sparse attention to cut long-context cost. Moonshot's own tables show leads on SWE Marathon (51.7) and HLE with tools (70.0), but trailing results on ProgramBench and the cybersecurity exploit tests. Its hallucination benchmark is internal, and none of the figures has been independently verified. The API is OpenAI-compatible, and vLLM and SGLang are supported.
Why it ranked
No. 20 of 139 with a score of 0.69. The panel rated it highest on new information (0.93) and reader interest (0.80), lowest on audience breadth (0.75). The judges disagreed most on promotional (spread 0.45), so the story is marked contested. Placed No. 20, No. 22 by score alone: the fifth AI & machine learning story of the day (−0.040). The placement terms move stories; they are not part of the score.
- AI judgeJudge 2claude-haiku-5-5
- Major open-weight 2.8T MoE release with detailed architecture and extensive benchmarks; highly relevant to builders, though many scores are self-reported and internal benchmarks limit verification.
- AI judgeJudge 3gpt-5.6-luna
- A major open-weight frontier model release with extensive architecture, benchmark, deployment, and usage details. Its practical relevance is high, though evidence is largely vendor-reported and the comparative capability framing is somewhat promotional.
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
0.69total under rubric v2.3Contestedjudges differ by 0.45 on promotional
- Significance0.75
- Reader interest0.80
- Audience breadth0.75
- Practical relevance0.80
- Substantiated0.76
- Technical depth0.80
- New information0.93
- Counts against: Promotional0.30
- Counts against: Overclaim0.35
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? | 0.71 | 0.72 | 0.82 | 0.75 | 0.11 | +0.24 | +0.180 |
| Reader interestWould a well-informed senior engineer or tech founder want to read this today, and likely pass it on? | 0.63 | 0.85 | 0.93 | 0.80 | 0.30 | +0.26 | +0.209 |
| Audience breadthHow many people who build technology does it concern? | 0.62 | 0.80 | 0.82 | 0.75 | 0.20 | +0.13 | +0.097 |
| Practical relevanceHow useful is it to someone who builds software or tech products? | 0.82 | 0.80 | 0.78 | 0.80 | 0.04 | +0.22 | +0.176 |
| 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. | 0.97 | 0.60 | 0.72 | 0.76 | 0.37 | +0.05 | +0.038 |
| Technical depthHow technically substantive is it? | 0.92 | 0.70 | 0.78 | 0.80 | 0.22 | +0.05 | +0.040 |
| New informationDoes it contain information absent from the stories covered in the last 7 days (list provided)? | 0.91 | 0.90 | 0.99 | 0.93 | 0.09 | +0.05 | +0.047 |
| 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. | 0.13 | 0.20 | 0.58 | 0.30 | 0.45 , above the contested threshold | −0.19 | −0.058 |
| OverclaimDo the headline or conclusions claim more than the body supports? | 0.29 | 0.35 | 0.42 | 0.35 | 0.13 | −0.11 | −0.039 |
| Sum of contributions | +0.690 | ||||||
| Claim-check penalty | 0.000 | ||||||
| Total, as published | 0.690 | ||||||
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/2026-11-05/20/"><img src="https://rubric.hrih.si/badge/2026-11-05/20.svg" alt="Ranked #20 on Rubric, 5 Nov 2026" width="238" height="28"></a>Claim check
Claim check: 6 no source given · 2 couldn’t check. Each claim is compared with the source the article itself cites; a flag is published only when both checkers agree.
No source givenNumber
Kimi K3.1 Pro has 2.8T total parameters with 104B activated parameters in a Mixture-of-Experts architecture.
The article gives no link or citation for it.
The article says “Total Parameters 2.8T”
The source says No source to compare with.
No source givenVersion
Kimi K3.1 Pro is the first model in the K3.1 series, post-trained from Kimi K3 with large-scale joint reinforcement learning across hundreds of downstream domains.
The article gives no link or citation for it.
The article says “Post-trained from Kimi K3 with large-scale joint reinforcement learning across hundreds of downstream domains, it is the first model in the K3.1 series”
The source says No source to compare with.
No source givenGeneralization
Kimi K3.1 Pro has a 1-million-token context window and delivers substantial improvements in coding, 3D creation, financial research, factual accuracy and other agentic capabilities.
The article gives no link or citation for it.
The article says “With a 1-million-token context window, Kimi K3.1 Pro delivers substantial improvements in coding, 3D creation, financial research, factual accuracy, and other agentic capabilities”
The source says No source to compare with.
Couldn’t checkBenchmark result
Kimi K3.1 Pro scores 65.7 on Terminal-Bench 4.0, compared with 66.4 for Claude Opus 5.5 and 42.3 for DeepSeek-V4.1-Flash-1017.
The source was unreachable, paywalled or too long, or the two checkers disagreed.
The article says “Terminal-Bench 4.0 65.7 57.9 58.2 66.4 42.3 41.8”
The source says No passage could be compared.
Source cited by the article: tbench.ai
Couldn’t checkBenchmark result
Kimi K3.1 Pro scores 51.7 on SWE Marathon v1.1, the highest among the compared models.
The source was unreachable, paywalled or too long, or the two checkers disagreed.
The article says “SWE Marathon v1.1 51.7 45.6 42.5 46.3 40.0 42.5”
The source says No passage could be compared.
Source cited by the article: swe-marathon.org
No source givenBenchmark result
Kimi K3.1 Pro scores 70.0 on HLE-Full with tools, ahead of Claude Opus 5.5 at 67.7 and Claude Fable 5.1 at 65.6.
The article gives no link or citation for it.
The article says “HLE-Full w/ tools 70.0 65.6 57.2 67.7 58.8 62.5”
The source says No source to compare with.
No source givenBenchmark result
Kimi K3.1 Pro has a hallucination rate of 8.3 on Kimi Hallu Bench Hard with tools (lower is better), versus 23.3 for Claude Fable 5.1 and 15.0 for Claude Opus 5.5.
The article gives no link or citation for it.
The article says “Kimi Hallu Bench Hard w/ tools (Internal, lower is better) 8.3 23.3 5.5 15.0 23.3 22.6”
The source says No source to compare with.
No source givenCausal conclusion
Kimi K3.1 Pro replaces the full-attention Gated MLA layers of Kimi K3 with Gated DeepSeek Sparse Attention layers for more efficient long-context modeling.
The article gives no link or citation for it.
The article says “Compared with Kimi K3, we replace the full-attention Gated MLA layers with Gated DeepSeek Sparse Attention (Gated DSA) layers for more efficient long-context modeling”
The source says No source to compare with.
Also covered by
No other coverage was clustered with this story.
Elsewhere
Elsewhere: HN 447 pts, front page
Recorded for context only, checked 5 Nov 2026, 05:46 UTC. It played no part in the ranking.