Kimi K3.1 Pro arrived on Venice on 23 October, with a 20% discount on Pro and Highspeed until the 31st. It is the strongest model the platform has ever offered. But the real story is not the model: it is the vocabulary it is sold with, which always runs a little ahead of the artefact.

What’s actually inside

Moonshot describes it as an open-weight model of the 3-trillion-parameter class: 2.8 trillion total, 104 billion active per token, mixture of experts, one-million-token context and 128K output. On Terminal-Bench 4.0 it scores 65.7%, against 66.4% for Claude Opus 5.5 and 58.2% for GPT-6 Astra: it does not beat them, it breathes down their neck at a third of the price. On Venice, access costs $3 per million input and $15 output, with cached input at $0.30.

The delicate point is the word “open”. The weights of K3.1 Pro are not public: Moonshot has promised them by 4 November. Until that day it is a closed model like the others, reachable only through the API and the app. The promise carries a stated price: Moonshot stripped cyber data and tasks from the training corpus and from the reinforcement learning mix, so as not to “extend the absolute frontier of cyber capability”. On its internal benchmark the public version scores 81% against 90.5% for the unfiltered one. It is a reasonable choice and should be said without irony: an open weight is a weapon anyone can sharpen.

Anonymity is a policy, not an enclave

Venice presents it as anonymous, and it is: prompts are not stored, profiled or used for training. The model page, however, spells out what the post on X does not: for this model there is no TEE and no end-to-end encryption. The guarantee is the anonymisation policy, not the hardware. Anyone using Venice for reasons of confidentiality should read that line twice.

The market isn’t celebrating

VVV is worth between $24.4 and $24.7, with a market cap around $1.2 billion and volume near $27-28 million. The structural picture is the same as every week: protocol TVL is zero under DefiLlama’s methodology, which keeps staking as a separate series, and $853 million is staked, 71.1% of the market cap. On-chain fees over the last thirty days are $997,888, $3.84 million cumulative, and they only cover VVV buybacks and burns: subscriptions, API and credits settle off-chain.

The question that remains

Why should a model that is “open” without being open yet and “anonymous” without passing through an enclave be of interest? For a simple reason: the gap between a written promise and a kept one gets measured on 4 November. Until then, what Venice is actually selling is a price — a million tokens of context at a third of the rival’s — and the chance to use it without an account. That is not nothing. It is also not what the label says.

Verified sources: @AskVenice on X, 23/10/2026 (availability, anonymous form, 20% discount until 31/10); Venice model page, venice.ai/models/kimi-k3-1-pro, consulted on 24/10/2026 (1M context, 128K output, price $3/$15 per million, cache $0.30, absence of TEE and end-to-end encryption for this model); @Kimi_Moonshot on X, 23/10/2026 (2.8T total, 104B active, weights by 4/11); Reuters via GalvNews, 23/10/2026 and Teck Hustlers, 23/10/2026 (removal of cyber data, xdev-bench-hard 81% against 90.5%, Terminal-Bench 65.7% against 66.4% and 58.2%); DefiLlama, defillama.com/protocol/venice, reading of 24/10/2026 (TVL $0, staking $853.19 million, 71.14% of mcap, 30d fees $997,888, $3.84 million cumulative); CoinGecko, Venice Token page, 24/10/2026 ($24.41). Discrepancies noted: Echo’s report put VVV at $24.79 with −4.4% over 24 hours; CoinGecko and DefiLlama read $24.41-24.71 with disagreeing changes, so the draft omits the percentage. Manufacturer metrics (2.8T/104B, 65.7%) are Moonshot claims reported by trade press, not independent measurements. The $1.7 billion TVL cited in the report was not used: it does not exist on DefiLlama.