Coverage

News

Google turns off the three Veo 3.1 preview endpoints today, with Gemini Omni 1.1 Flash as the only named replacement

Developers calling Veo 3.1 through Google's consumer-facing Gemini API hit their deprecation date today. The migration target is a different model with a different call shape, and it is the second forced Veo move in four months.

news 22 Oct 2026

NVIDIA ships a real-time surgical video model and states that visual realism is not proof of accuracy

A video generation release that declines to claim its own output is correct. NVIDIA's new surgical world model streams interactively on one GPU, and the announcement attaches no frame rate, no clinical evaluation and no comparison against a conventional simulator.

news 22 Oct 2026

VideoGen unveils Model 1 with a ten-minute ceiling and no published price, benchmark or third-party number

A Y Combinator-backed company put a ten-minute number and a 50% cost reduction in the same press release, then listed pricing, benchmarks and outside validation as things that have not happened yet.

news 21 Oct 2026

DecisionPivot scores 40 automated judges across 40,033 items, and 19 of them land under a guessing baseline

The benchmark is built to show what aggregate accuracy hides in the small classifier models now sold as automated judges. Its 40,033 items are entirely text, which is the gap for anyone scoring generated video or images.

news 21 Oct 2026

FLUX 3 Max ships with three different timings for the same 10-second clip

A single headline claim, a 10-second clip generated in about 12 seconds, appears three different ways across the FLUX 3 Max launch. The quality parity it depends on comes from 4,868 arena votes.

news 20 Oct 2026

ByteDance puts video generation at about 10% of its compute, against a 55% token figure it calls unreliable

A viral claim said over half of China's AI compute is making short dramas. ByteDance's answer is a single percentage, measured a different way, from a different source, and nobody can check either one.

news 19 Oct 2026

NVIDIA's AnyFlow video checkpoints are on Hugging Face, where eight times the sampling steps buys 0.06 VBench points

AnyFlow's pitch is a single video model that gets better the more steps you give it. The table behind that pitch shows one variant gaining six hundredths of a VBench point for eight times the sampling.

news 19 Oct 2026

Black Forest Labs ships FLUX 3 Image Klein at 5.6B parameters, with every first-place claim scored in-house

A 5.6B open weights model that runs on 8 GB of VRAM and claims to beat 7B and 12.9B rivals. The scoreboard behind that claim belongs to Black Forest Labs, and the free commercial licence stops at $5 million in revenue.

news 18 Oct 2026

Auros asked 2,000 US shoppers to separate real products from AI ads, and published one accuracy number: 52%

Shoppers say they are being fooled by AI advertising, and 43% of Americans report feeling duped. When the same study actually tested identification, the published score for real products was 52%.

news 18 Oct 2026

Kling 4.0 launches, and the 4K 10-bit HDR it was announced with is still listed as coming soon

Kling 4.0 is live after three weeks of October promises, but the broadcast specs that sold it and the developer API are both still written in the future tense, and the two launch releases disagree on what the model does.

news 17 Oct 2026

OpenAI and Meta tell a US evidence-rules panel that no method reliably detects AI video

Two of the largest makers of generative image and video models spent a federal rulemaking meeting explaining that the output of those models cannot be reliably told apart from real footage, by people or by software.

news 16 Oct 2026

FlashVSR+ claims untiled 4K video upscaling on one 24GB RTX 4090, trained against a black-box reward model

A new open video upscaler says it can hit 4K on a consumer graphics card without tiling, and it gets its quality alignment from a reward model that is never differentiated through.

news 16 Oct 2026

Higgsfield's 10-minute AI video claim appears in an OpenAI customer story, with no sample and no availability detail

A ten-minute ceiling is a big number for AI video. The page carrying it is a vendor customer story, the quote is from the buyer's CEO, and the model credited for the jump does not render a single frame.

news 15 Oct 2026

xAI puts Grok's image and video generation into one weekly pool shared with a coding agent

Grok's generative media no longer has its own plan. It draws from the same weekly allowance as a coding agent, and the tier copy does not agree on which plan includes video at all.

news 15 Oct 2026

Artificial Analysis ships a Decision Index for the models being sold as cheap AI judges, and all four of its evaluations are text

The first independent scorecard for decision models landed on October 13. It ranks the exact class of model vendors are pitching as a replacement for LLM judges, using four evaluations that never show the model a picture.

news 14 Oct 2026

BudgetPixel's image arena separates its top two models by 6 Elo, and their intervals overlap by 14 points

A nightly-refit image leaderboard crowned a new number one on October 12, then printed the confidence intervals that show it cannot tell first from second. The fourth-place model got there on 665 votes.

news 13 Oct 2026

Synthesia turns on Syren for everyone, with a 1,000 to 2,000 credit range and a $5 claim it has not priced

Syren is live, in beta, and capped at three minutes per video. The arithmetic behind the "$20K agency videos for $5" pitch is the one thing Synthesia has not published.

news 11 Oct 2026

Grok Imagine Video 1.5 Lite enters the Artificial Analysis board at Elo 993, with a rank range wider than the gap being claimed

A distilled video model is being marketed on a $0.02 per second figure that applies to 480p, while the benchmarked tier runs $0.14 per second and upscales 720p to 1080p.