vLLM Delivers Day-0 Support for Kimi K3.1 Pro
Comments on “vLLM Delivers Day-0 Support for Kimi K3.1 Pro”
Claim-checked. Scored by all three judges, and its checkable claims were compared with the sources it cites. How far stories are checked
vLLM has added day-0 support for Moonshot AI's Kimi K3.1 Pro, a 2.8T-parameter open-weight multimodal MoE model with a 1M-token context. The main change is DSA sparse attention in its 24 full-attention layers, which cuts the memory reads per decode step at long context to roughly a quarter of dense…
Why it ranked
No. 22 of 139 with a score of 0.70. The panel rated it highest on technical depth (0.96) and new information (0.89), lowest on audience breadth (0.63). Placed No. 22, No. 14 by score alone: the sixth AI & machine learning story of the day (−0.060), and close to “Day 0 support for Mistral Large 4 on vLLM”, No. 4 on Friday 30 October 2026 (−0.014). The placement terms move stories; they are not part of the score.
- AI judgeJudge 2claude-haiku-5-5
- Deep, specific technical detail on sparse attention, KV cache and inference kernels for a major open-weight model; mostly a vendor release post with benchmarks, moderately promotional but substantive.
- AI judgeJudge 3gpt-5.6-luna
- A highly technical day-zero serving and training integration, with reproducible commands, architecture details, benchmarks, kernel-level optimizations, and multi-accelerator support. It is somewhat promotional, but unusually substantive and practically actionable for inference engineers.
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
0.70total under rubric v2.3
- Significance0.70
- Reader interest0.82
- Audience breadth0.63
- Practical relevance0.81
- Substantiated0.85
- Technical depth0.96
- New information0.89
- Counts against: Promotional0.21
- Counts against: Overclaim0.29
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? | 0.74 | 0.55 | 0.82 | 0.70 | 0.27 | +0.24 | +0.169 |
| Reader interestWould a well-informed senior engineer or tech founder want to read this today, and likely pass it on? | 0.80 | 0.70 | 0.95 | 0.82 | 0.25 | +0.26 | +0.212 |
| Audience breadthHow many people who build technology does it concern? | 0.56 | 0.60 | 0.72 | 0.63 | 0.16 | +0.13 | +0.081 |
| Practical relevanceHow useful is it to someone who builds software or tech products? | 0.84 | 0.70 | 0.88 | 0.81 | 0.18 | +0.22 | +0.177 |
| 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. | 0.86 | 0.80 | 0.90 | 0.85 | 0.10 | +0.05 | +0.043 |
| Technical depthHow technically substantive is it? | 0.99 | 0.92 | 0.98 | 0.96 | 0.07 | +0.05 | +0.048 |
| New informationDoes it contain information absent from the stories covered in the last 7 days (list provided)? | 0.84 | 0.85 | 0.98 | 0.89 | 0.14 | +0.05 | +0.045 |
| 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. | 0.09 | 0.35 | 0.18 | 0.21 | 0.26 | −0.19 | −0.039 |
| OverclaimDo the headline or conclusions claim more than the body supports? | 0.28 | 0.30 | 0.30 | 0.29 | 0.02 | −0.11 | −0.032 |
| Sum of contributions | +0.704 | ||||||
| Claim-check penalty | 0.000 | ||||||
| Total, as published | 0.704 | ||||||
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/2026-11-05/22/"><img src="https://rubric.hrih.si/badge/2026-11-05/22.svg" alt="Ranked #22 on Rubric, 5 Nov 2026" width="238" height="28"></a>Claim check
Claim check: 1 matches source · 5 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 givenBenchmark result
Kimi K3.1 Pro is a 2.8T-parameter (104B active) multimodal MoE with a 1M-token context that matches the strongest closed-source models on coding, agentic and knowledge benchmarks.
The article gives no link or citation for it.
The article says “a 2.8T-parameter (104B active) multimodal MoE with a 1M-token context that matches the strongest closed-source models on coding, agentic and knowledge benchmarks”
The source says No source to compare with.
No source givenBenchmark result
Token-sharded prefill attention gives up to 1.69× faster time to first token than head-sharded prefill at 340K tokens on 8× GB300 (15.89 s vs 9.42 s).
The article gives no link or citation for it.
The article says “up to 1.69× faster time to first token (TTFT) than head-sharded prefill at 340K tokens on 8× GB300”
The source says No source to compare with.
No source givenNumber
With DSA, at long context a decode step reads about 23% of the bytes per layer that dense MLA reads.
The article gives no link or citation for it.
The article says “At long context, that is about 23% of the bytes dense MLA reads per layer”
The source says No source to compare with.
Couldn’t checkBenchmark result
The DSpark preview speculator reaches a mean acceptance length of 4.597 on GSM8K with 7 draft tokens, giving a 3.0× decode speedup at batch size 1 with TP8 on 8× GB300.
The source was unreachable, paywalled or too long, or the two checkers disagreed.
The article says “the preview already reaches a mean acceptance length of 4.597 on GSM8K with 7 draft tokens. At batch size 1 with TP8 on 8× GB300, that is a 3.0× decode…”
The source says No passage could be compared.
Source cited by the article: huggingface.co/Inferact/Kimi-K3.1-Pro-DSpark-preview
No source givenBenchmark result
On NVIDIA GB300 on the SemiAnalysis AgentX benchmark, Kimi K3.1 Pro reaches 375.9 tok/s per user at the latency-sensitive end and 37.8K total tok/s per GPU at the throughput end.
The article gives no link or citation for it.
The article says “on NVIDIA GB300, up to 375.9 tok/s per user at the latency-sensitive end and 37.8K total tok/s per GPU at the throughput end”
The source says No source to compare with.
No source givenBenchmark result
On NVIDIA Rubin, Kimi K3.1 Pro reaches 815.6 tok/s per user and 68.7K total tok/s per GPU on AgentX.
The article gives no link or citation for it.
The article says “on NVIDIA Rubin, up to 815.6 tok/s per user and 68.7K total tok/s per GPU on the SemiAnalysis AgentX benchmark”
The source says No source to compare with.
Matches sourceFirst, only or fastest
For the first time, vLLM day-0 support ships zero-mismatch RL, with bit-identical log-probs between vLLM and Megatron so the rollout/trainer KL is exactly 0 across a 100-step full-model run.
The cited source says the same thing.
The article says “for the first time, day-0 support comes with zero-mismatch RL: vLLM and Megatron produce bit-identical log-probs for both the base model and the LoRA adapter”
The source says “100 DAPO steps with LoRA, 0 mismatch at every step; AIME 24/25/26 mean@8 0.451 → 0.664 (16 GB300”
Source cited by the article: github.com/Inferact/zero-mismatch-rl
Couldn’t checkFirst, only or fastest
The Zhenwu PPU's support for Kimi K3.1 Pro is the first time it offers day-0 support for a mainstream open-source large model.
The source was unreachable, paywalled or too long, or the two checkers disagreed.
The article says “This marks the first time the Zhenwu PPU offers Day-0 support for a mainstream open-source large model”
The source says No passage could be compared.
Source cited by the article: github.com/vllm-project/vllm-sail
Also covered by
No other coverage was clustered with this story.
Elsewhere
Elsewhere: HN —
Recorded for context only, checked 5 Nov 2026, 05:46 UTC. It played no part in the ranking.