Skip to content
World clockEU--:--UK--:--USA--:--CN--:--PLDEFRIT中文EN

portal about AI and technologyevents · analysis · interviews · technical background

Search
LIVE
›

TBC's mischief 1 claims 5x faster AI video at half a cent per second

The Biological Computing Co. released TBC mischief 1 on 30 October 2026, an image-to-video model it says generates video five times faster than its base model, Wan 2.2, at half a cent per second of output.

AI & modelsNewsGrace OkonkwoPublished: 4 November 20265 min readSources 28
TBC's mischief 1 claims 5x faster AI video at half a cent per second

TBC, as the San Francisco company is known, built mischief 1 on Wan 2.2, an open-weight video model. It says the result matches Wan 2.2's quality while generating video five times faster.

A fivefold speed claim, at half a cent per second

Pricing starts at half a cent per second of output, which puts a five-second clip at 2.5 cents and 100 clips at $2.50. The model is live on fal and OpenRouter, and TBC says it is the lowest-priced image-to-video model on both marketplaces.

The company describes the gain as a thin optimization layer rather than a new foundation model. It adds less than 0.1% to the base model's parameter count, and customers run it on standard AI hardware. Output is modest, at 3 or 5 second clips in 480p or 720p, 16 frames per second, delivered as MP4.

Cheap per-second pricing is the pitch. Independent evidence for the speed claim is not yet public.

"We said we would take discoveries from living neurons all the way to a product anyone can use, and we did it faster than anyone expected," said Alex Ksendzovsky, TBC's CEO and co-founder.

What the neurons did, and what the evidence shows

TBC grows living neurons in the lab, measures how they process information and turns the principles it finds into software. The neurons stay in the lab; what ships is code that runs on ordinary accelerators.

The framing has invited confusion, and the coverage is consistent on one point: nobody is generating video on a dish of brain cells. What TBC sells is an efficiency layer on top of an existing model, and the biological work sits upstream. TNW pointed out that TBC has not published independent tests backing the fivefold speed claim, a gap that shapes how the launch should be read, shattered.io noted. An AWS Marketplace listing is marked coming soon, following the AWS partnership TBC announced on 22 September 2026. No funding amount or investor list appears in the launch materials.

Jon Pomeraniec, TBC's co-founder and COO, framed the launch as a proof of method. "Our first model shows what happens when we learn from biology instead of relying on conventional inference optimization alone," he said.

Wonder Studios, an AI-native creative studio, tested the model in its own production workflow. "mischief 1's speed and cost-effectiveness meant we could push the work much further within the same time and more cost effectively," said co-founder Justin Hackney in TBC's announcement.

Free credits and cheaper APIs widen the field

The same week brought cheaper access elsewhere. AIVideoFree said on 30 October that new users' sign-up credits work on all 12 of its video models, including Veo, Kling and Seedance. Videos carry no watermark, credits do not expire and commercial use is included.

Extra credits sell in one-time packs from $6.90 for 600 credits, and annual plans start at $7.40 per month, billed yearly. No credit card is required, though the number of free credits varies by country, and the credit cost of each generation is shown before it starts. On 28 October, B.AI added ByteDance's Seedance series to its API. Seedance 2.5 handles runs up to 30 seconds at 1080p 10-bit and up to 50 reference assets. Meta gives away an AI video generator inside its Meta AI app and on meta.ai. It runs on Movie Gen, a 30-billion-parameter model that makes 1080p clips up to 16 seconds at 16 frames per second.

Meta's free tier comes with a feed attached. Vibes launched on 25 September 2025, and Meta AI app downloads rose 56% month over month to about 3.9 million by mid-October 2025, Fello AI reported.

AIVideoFree still restricts content: material involving minors, sexual or nude depictions of real people without consent, impersonation for fraud and illegal content are not permitted, and some models apply their providers' own safety filters.

What buyers can compare

Specifications diverge sharply. TBC's 3 to 5 second clips at 480p or 720p sit below Meta's 1080p, 16-second Movie Gen output and below Seedance 2.0's 4K 10-bit support.

AIVideoFree's free tier spans 12 video models, while TBC sells a single model through two marketplaces. Comparing prices directly is awkward. TBC quotes half a cent per second of output, while AIVideoFree sells 600 credits for $6.90 without publishing a per-second rate, so the two meter usage differently. B.AI's Seedance 2.0 Fast targets the same low-resolution, high-volume work, generating at 480p or 720p for ad variants and storyboard iteration.

Buyers who need volume and price can test more takes per dollar. Buyers who need resolution and length still have to look elsewhere.

What the market has not settled is whether speed claims survive contact with production workloads.

What happens next

TBC says it is already applying its discovery process to newer open-weight video models, with early results showing greater efficiency gains. It has also said it is working on language models and physical AI.

The AWS Marketplace listing for mischief 1 is marked coming soon, with no date given. Until independent benchmarks appear, the fivefold figure remains a vendor number.

Comments 0

Sources

28
  1. 01Neuron-Derived AI Video Model: 5x Faster Claim [2026]EN
  2. 02The Biological Computing Co. Brings Neuron-Derived AI Video Model to MarketEN
  3. 03AIVideoFree Relaunches as a Free AI Video Generator With No Restrictions on ModelsEN
  4. 04ByteDance Seedance Series Video Generation Models Launch on B.AI APIEN
  5. 05Meta AI Video Generator: How It Works & Is It Free (2026)EN
  6. 06AI Carbon Footprint: What One Prompt, Image, Video and Model Really Emit - Fello AIEN
  7. 07China touts AI milestone: models generate more tokens than used in trainingEN
  8. 08BYD teases new model for Guangzhou Auto Show, with next-gen Han expectedEN
  9. 09Anthropic is working on a system to prove AI models’ identityEN
  10. 10Nvidia is happy to take your money whether you're using open or closed models, says Jensen HuangEN
  11. 11STAT+: Amid Mass General Brigham layoffs, CEO pens vague letter, stirring staff anxietyEN
  12. 12Aluminum–Air Generators for Data Centres Face a Metal Logistics TestEN
  13. 13 Google reveals new Gemini 4 Carbon AI model, narrowly beating Claude Opus 5.5 EN
  14. 14Uwear.ai – AI Fashion Visual Production, On-Model Photos & Video Generation - Quasa.ioEN
  15. 15LFM3, a suite of small, high-performance AI models that run on smartphones, has been released, including next-generation models such as language models, visual EN
  16. 16I Checked Out Tesla’s New Model YL, Standard Y, Model 3 In Person — Here’s What Surprised MeEN
  17. 17heise+ | Humanoide Roboter in Deutschland: Modelle, Preise und VerfügbarkeitEN
  18. 18Behind-the-meter generation could deliver five times more grid relief than data center load shiftingEN
  19. 19Open models can do 90% of agentic work at a twentieth of the cost of a frontier LLM: GartnerEN
  20. 20Tutor Intelligence raises Series B to scale model training, robot deploymentsEN
  21. 21Jülich Hosts EVE Language Model for Earth Observation and Earth ScienceEN
  22. 22Google Launches LiteRT Tensor API for Building, Customizing, and Optimizing ML Models for On-DeviceEN
  23. 23Westward Ho! First Trip Cross Country with Our New 2026 Tesla Model Y & Full Self Driving V14EN
  24. 24AI, la sfida degli specialisti: oltre i generalisti c’è spazio anche per l’ItaliaEN
  25. 25Red Hat Patches 9.8-Rated Linux TCP RCE Flaw While Ubuntu Releases Fixes for TestingEN
  26. 26RTL-Interact: Training EDA-Interactive Models for Multi-Task RTL DesignEN
  27. 27V0.5.0 of PSSA: a non-transformer language modelEN
  28. 28The Share of Solar Generation with Co-located Battery Storage Is IncreasingEN

All figures and quotations in this text come from the sources listed below.

Content prepared by the editorial team with AI assistance.

Grace Okonkwo

Grace Okonkwo

AI, models and technology

Grace Okonkwo covers AI, models and technology for FLASH24, working from primary sources such as model cards, API documentation and benchmark papers rather than vendor summaries. She checks training data provenance, evaluation conditions and reported scores against the underlying datasets before any figure reaches print. She interviews researchers and engineers directly, tracks release calendars from major labs, and compares successive model versions on the same tests. Her own self-hosting, home-network and documentation-reading habits feed straight into that desk, since she tests tools on her own hardware first. She does not publish benchmark claims without a reproducible method.

Newsroom →

Comments

0
  1. No comments yet — be the first.

Write a comment

Comments are public. We do not publish abuse, spam or advertising.