The number is round and it comes from the official account: five million users, announced on 19 October. It is a threshold worth taking seriously, with the caveat that applies to every metric the company publishes: declared, not certified. An indirect confirmation sits in the Lumara Film Festival release, presented by MoonPay and Venice, which describes the platform as “used by 5 million users”. In July the 65 million Series A at a one billion valuation spoke of 3.5 million accounts.

The model that doesn’t write one word at a time

The technical piece is Mercury 3, a diffusion model from Inception Labs, released on 19 October. Instead of one token at a time, it refines blocks of positions in parallel: 1,068 tokens per second on NVIDIA GPUs, a one-million context, more than forty times the size of Mercury 1 at comparable speed. Inception claims it beats GPT-6 Luna and Claude Haiku 5.5 while running four to eight times faster; Venice’s post compressed that range into two separate figures — four times Haiku, eight times GPT-6 — which is a commercial simplification of the manufacturer’s number. On Venice it runs in the anonymous tier, billed per token (0.10 and 0.35 dollars per million).

The point is not the leaderboard, self-referential until a third party shows up: it is that a non-autoregressive architecture enters a consumer catalogue on launch day. If it holds, the latency that piles up at every step of an agent shifts.

A catalogue that fills up in a day

In the same window came MiniMax M3.1 Flash, one-million-token context, images and video as input, presented as three times faster than M3; and FLUX 3 Max, video post-trained by fal Research on the FLUX 3 base, clips up to twenty seconds in 1080p with native audio and ten seconds generated in about eleven. Manufacturer numbers, these too.

The catalogue is the product, and the cultural shopfront follows: the Lumara Film Festival received over 700 submissions and announces its winners on 26 October in New York.

The token tells a different story

VVV trades around 23 dollars according to CoinGecko (23.22, up 5.6% in 24 hours) and 24.5 according to Bybit. The figure that matters is another one: according to DefiLlama about 452 million dollars of VVV are staked, 71.4% of the market cap. The “TVL” field stays at zero, and that is not an error: the protocol has no liquidity in DeFi pools and staking is counted separately. It needs saying, because the radar data carried a “Venice TVL” of 1.74 billion that does not exist on DefiLlama.

The W42 weekly report from a third-party account, self-declared as AI-generated, tells another week: VVV down 8% at 22.77 dollars, but 224,000 net VVV more in staking and signups at 21,935. Figures Venice has not confirmed. If they hold, they say the most interesting thing: users and staking do not follow the price. Five million people are a measure of distribution, not of revenue.

Verified sources: @AskVenice post on X, 19 October 2026 (five million users); Lumara Film Festival release via Yahoo Finance, 16 October 2026 (“trusted by 5 million users”); venice.ai/models/mercury-3 and Inception Labs blog “Introducing Mercury 3”, 19 October 2026; venice.ai/models/minimax-m3-1-flash and venice.ai/models/flux-3-max (October 2026); CoinGecko and Bybit for the VVV price, consulted on 20 October 2026; DefiLlama, Venice protocol page, consulted on 20 October 2026 (staking 452.27 m, 71.38% of mcap, TVL 0 $ by methodology); Unite.AI for the July 2026 Series A. Discrepancy noted: the “Venice TVL 1,740,314,272 $” in Echo’s report is not confirmed by DefiLlama, which reports TVL at zero and staking as a separate series; the verified figure is used in the text.