News
Same AI model, 48 prices: the weekly AI API price board
GLM-5.3 costs $0.40 a million tokens at Wafer and $4.30 at Alibaba. Every model rate, every price move and every outage, rebuilt every Tuesday.
Latest
Everything we've published, newest first.
H3 Max
H3 Max Relight changes a video's lighting from a sphere render
The fal endpoint takes a clip of up to 15 seconds plus a lighting-sphere render and bills from 5 cents a second at 480p.
WorldPlay2
WorldPlay2 turns one image into a playable browser world
The weights landed on October 6 under a non-commercial license, and a hosted version went live on Reactor the next day.
SGF+
Tsinghua's SGF+ generates 24-hour video from 5-second clips
The method doubles generator parameters to 2.8 billion to separate the two roles, and its weights and code ship under Apache-2.0.
FreeVideo
How to run MiniMax H3 on an 8 GB graphics card
FlashML-org's FreeVideo runs an eight-step MiniMax H3 derivative on small cards by streaming weights between GPU, RAM and disk.
GRACE
KAIST's GRACE makes Wan2.1 video 11x faster on an A100
Two-stage latent compression cuts Wan2.1-I2V-14B's token count nearly 8x while matching the model's VBench quality on one A100.
AI video API
The 8 cheapest AI video APIs, ranked by price per second
Vidu Q4 Preview leads at $0.014 a second on launch pricing, while Google's Veo 3.1 Lite, Alibaba's Wan 3.0 and MiniMax's H3-Max tie at five cents.
ai-creative-summit
AI Creative Summit 2026 offers 30 free tickets for freelancers
The fourth AI Creative Summit runs at BFI Southbank in London on 11 November 2026, with a keynote from Tilly Norwood creator Eline Van der Velden.
Kling
Report: Kling AI picks banks for $1 billion Hong Kong IPO
Kuaishou's Kling also unveiled its Kling 4.0 model on September 28, doubling single-pass clips to 30 seconds and previewing up to 10 keyframes.
Runway
Runway's Praxis-1 turns video pretraining into robot control
The New York company says early partners Noble Machines, Standard Bots and Ultra are running Praxis-1 on their own hardware, with open weights planned for the coming months.
Vidu
Vidu Q4 Preview sells AI video from $0.014 a second at launch
Shengshu's launch rate is promotional: Vidu's own list starts at $0.045 a second, and the model debuts third on an independent image-to-video board.
Fable
Gods Don't Give Gifts moves to December 4 for Oscar bid
Its trailer debuts in theaters on October 9 and a Los Angeles billboard push follows in November, as Fable submits the film for Best Animated Feature.
SynthID
How to check media for AI with Google's SynthID Detector
Google opened its watermark checker to the public on October 7, but a clean result means no SynthID mark was found, not that a file is real.
higgsfield
Higgsfield builds an AI influencer character for $0.06
The rebuilt tool assembles a character from 19 menu settings with no prompt writing, takes your own photo to carry your face, and animates it through Genjutsu.
Spira Maxima
Spira Maxima turns a script into a finished social video
Spira AI says the model bakes a presenter, B-roll, captions and music into one pass, and that its agents can post it to TikTok, Instagram and X.
Astana AI Film Festival
Astana AI Film Festival awards $450,000 to a Latvian director
Organisers say 8,067 films from 125 countries made it the largest AI film competition yet, with $1 million split across eight winners.
invideo
invideo's Agentic Extend lengthens a scene by up to 30 seconds
invideo says the extension continues a scene's dialogue, music and sound effects on separate layers the editor can still adjust by hand.
xAI
Eighth Circuit pauses Minnesota's AI nudification ban
The St. Louis appeals court froze the first-in-the-nation law in a one-sentence order on October 2, while xAI's First Amendment challenge continues.
TwelveLabs
TwelveLabs Pegasus 1.6 reads first-person video at $1.75 an hour
TwelveLabs' Pegasus 1.6 adds first-person footage support for robot training data, at the same per-hour and token rates as the version it replaces.
Prism
Tencent Hunyuan's Prism renders 2K video with sound in one pass
The Fudan and Tencent Hunyuan model is MIT-licensed and open on GitHub, but native 2K output needs four or more 80 GB GPUs.