The neurons stay in the lab
The Biological Computing Co. (TBC) describes mischief 1 as a video model developed using findings from experiments with living neurons. That does not mean brain cells generate its video frames. The neurons remain in TBC's laboratory, where researchers stimulate cultured cells, record their responses and use the measurements to guide software development.
The commercial model runs on conventional AI hardware. TBC says it builds on Alibaba's open-weight Wan 2.2 and adds software accounting for less than 0.1% of the base model's parameters.
The distinction matters. TBC is proposing a biological source for AI optimization techniques, not replacing GPUs with living cells. Its published research provides examples of that approach, but the company has not disclosed enough about mischief 1's implementation to independently establish how much of its reported efficiency gain comes from neuron-derived methods rather than other optimizations.
What half a cent per second buys
OpenRouter and fal list mischief 1 at $0.005 per generated second for 480p output and $0.01 per second for 720p. At those published rates, a five-second 480p clip costs $0.025, while a five-second 720p clip costs $0.05. Generating 100 five-second clips at 480p would cost $2.50 in listed generation charges, before any other applicable costs.
The available configuration is narrow: image-to-video generation from an opening frame, with three- or five-second clips at 480p or 720p. OpenRouter lists a 16:9 aspect ratio. Its listing does not offer a separate text-to-video mode or vertical output.
That makes mischief 1 a candidate for inexpensive shot experiments, moodboards and previsualization rather than a direct substitute for longer or higher-resolution video models. The listed price is independently checkable; the quality and reliability of the results still need testing. Marketplace prices and available settings may change.
Five times faster, according to TBC
TBC says mischief 1 generates video five times faster than its Wan 2.2 base while maintaining comparable quality. Related company material also describes an 80% reduction in inference costs. Those are company-reported comparisons, not independently established performance results.
The October 30 announcement does not provide a reproducible benchmark protocol covering hardware, sampling settings, generation duration, quality measurements and cost calculations. It is therefore not possible to determine whether the reported speed advantage would hold under comparable third-party conditions.
There is also a distinction between TBC's published research and this commercial release. Its Neural Dynamics Adapter experiments used cultured-neuron measurements from a 4,096-electrode array and were demonstrated with Oasis, an interactive Minecraft-style video model. Mischief 1 instead uses Wan 2.2. TBC has not published an equivalent technical description of the shipped implementation.
The earlier research helps explain the proposed method. It does not independently validate mischief 1's fivefold speed claim or isolate the contribution of the biological measurements.
An inexpensive model to test, with an unanswered technical question
The low listed price makes mischief 1 relatively accessible for hands-on evaluation. Creators can test whether its short image-to-video clips meet their needs without committing to a large generation budget.
Testing TBC's performance claims requires more than trying the model in isolation. A useful comparison would run mischief 1 and the relevant Wan 2.2 baseline with matching prompts, source images, output settings and measured generation times, then assess quality under the same criteria.
Until comparable results are published, the defensible conclusion is limited: mischief 1 is a publicly accessible Wan 2.2-based video model with low listed generation prices and an unusual research origin. Its claimed speed advantage and the precise contribution of its neuron-derived techniques remain unverified.
Sources
- [1] The Biological Computing Co., “The Biological Computing Co. Launches World’s First Publicly Available Neuron-Derived AI Video Model”, press release, October 30, 2026, read first-hand: the launch date and San Francisco dateline, the Wan 2.2 base, the claim that mischief 1 generates video 5x faster than its base model while matching its quality, the half-cent-per-second pricing (a five-second clip at 2.5 cents; 100 clips at $2.50), the method statement (“The neurons stay in the lab”; software on standard AI hardware adding less than 0.1% to the base model's parameter count), the Wonder Studios production test and Justin Hackney quote, and the AWS collaboration announced five weeks earlierRead source
- [2] OpenRouter, mischief 1 model listing, read first-hand on November 1, 2026: $0.005 per second at 480p and $0.01 per second at 720p, 480p and 720p output, 3- and 5-second clips, 16:9 aspect ratio, steering by a first-frame image, image-to-video, released October 30, 2026Read source
- [3] fal, mischief 1 image-to-video listing and API documentation, read first-hand on November 1, 2026: the image-to-video endpoint requiring a prompt and a first-frame image, 480p (832x480) and 720p (1280x720) landscape output, 3- and 5-second durations, and published pricing matching OpenRouter at $0.005 per second (480p) and $0.01 per second (720p)Read source
- [4] Unite.AI, “TBC Launches Mischief, a Neuron-Derived AI Video Model Built on Wan 2.2”, launch coverage, October 2026: the absence of an independent benchmark or reproducible test detail in the announcement, and the Neural Dynamics Adapter research detail (cultured neurons recorded on a 4,096-electrode array; the adapter demonstrated on Oasis, an interactive Minecraft video model, rather than on Wan 2.2)Read source