MetaPatchAI home
AI

A Video Model Tuned by Lab-Grown Neurons? TBC's "mischief 1" Launches at Half a Cent per Second

Compiled by MetaPatchAI · Published , 3:25 PM ET

Oct 30, 2026: The Biological Computing Co. (TBC), a San Francisco startup that grows living neurons in a lab to find ways to make AI more efficient, today launched TBC mischief 1. It's an image-to-video model the company says generates video 5x faster than its base model. It's live now on fal and OpenRouter, and TBC says it's the lowest-priced image-to-video model on both marketplaces.

The name is a nod to Mischief Night: TBC says the Oct 30 launch date was chosen "to mark Mischief Night."

What mischief 1 is

According to TBC's press release, mischief 1 is built on Wan 2.2, an open-weight video model. TBC says it "matches its quality while generating video 5x faster and at a fraction of the cost."

The details from the model listings on OpenRouter and fal:

  • Input: an image plus a text prompt. The first frame image steers the video.
  • Output: video at 480p or 720p, in 3- or 5-second clips, at 16:9.
  • Price: from $0.005 per second at 480p and $0.01 per second at 720p. On fal's listing, a 5-second clip costs $0.025 at 480p or $0.05 at 720p.
  • Release date: OpenRouter lists it as released Oct 30, 2026.

TBC's headline line, "half a cent per second," means a five-second clip costs 2.5 cents. Note that this is the 480p price. 720p costs twice as much.

The "neuron-derived" part, explained

This is the unusual part. TBC says its discovery platform:

  1. Grows living neurons in the lab and measures how they process information.
  2. Identifies computational principles from those measurements that could improve AI.
  3. Turns those findings into proprietary software that makes existing models more efficient.

TBC says the neurons "stay in the lab." Customers run ordinary software on standard AI hardware, and TBC says it adds less than 0.1% to the base model's parameter count. The company argues that most AI efficiency gains come from shared techniques such as quantization, distillation and caching. Its own optimizations, it says, are "derived from biology, not just inspired by it."

Treat this as the company's description. TBC hasn't published the specific neuron-derived techniques in the sources we checked, and we haven't seen an independent benchmark of the 5x speed claim.

Who it's for

TBC is pitching mischief 1 at creators and developers who need lots of variations quickly. OpenRouter's listing says it's "ideal for ideation, previsualization, moodboarding and variant testing." TBC's press release quotes Justin Hackney, co-founder of Wonder Studios, an AI-native studio that tested the model. Hackney says its speed and cost "meant we could push the work much further within the same time."

TBC says this launch follows a collaboration with Amazon Web Services to bring the model to enterprise customers, announced "five weeks" earlier. It also says it's already applying the same process to "newer open-weight video models," with early results showing "greater efficiency gains." Work on language models and physical AI is in development.

Should you try it?

At these prices, testing is cheap: compare it with Wan 2.2 on the same input. Know this first: OpenRouter currently routes the model through one provider (Ambient), and it showed 77.34% availability over the last three days when we checked on Oct 30, 2026 (OpenRouter uptime page). Expect some failed requests while it ramps up.

What's confirmed / not confirmed

Confirmed (TBC press release, plus the OpenRouter and fal model listings): - mischief 1 launched Oct 30, 2026, on fal and OpenRouter. - It's an image-to-video model built on Wan 2.2, outputting 480p or 720p in 3- or 5-second clips. - Pricing is $0.005/sec at 480p and $0.01/sec at 720p. - TBC says its added software is less than 0.1% of the base model's parameters.

Company claims (not independently verified): - 5x faster than the base model, at matching quality. - "Lowest-priced image-to-video model" on both marketplaces. - The efficiencies come from measurements of living neurons.

Not confirmed: - Independent quality and speed benchmarks. - Technical details of the neuron-derived methods. - Timelines for the language-model and physical-AI versions.

Sources

Related

Up: AI