AI research lab - est. 2026

Small models that reason harder.

DeletedAI is a small AI lab working on efficient models and testing practical assistants, with a focus on usefulness and manageable costs.

Explore models Read the roadmap

Overview

DeletedAI is an AI lab created in 2026.

We are currently iterating and testing DeletedAI Mini at small scale, and we plan to train roughly 27B and 320B parameter models once funding reaches $30K.

Every model starts from a strong open base and is then tuned for efficient reasoning: shorter chains of thought, plainer language, and a sharper read on what the user actually asked for.

DeletedAI Mini
0.8B Iterating and testing small scale
Target
27B At $30K funding
Target
320B At $30K funding
Funding
$20 Current total

Model families

Three sizes, one goal: an assistant cheap and pleasant to work with

DeletedAI Mini

Iterating and testing small scale

Quicker reasoning and more concise answers.

Base
Qwen 3.5 0.8B
Parameter size
0.8B

DeletedAI Flash

Planned at $30K funding

More efficient reasoning, easier to read responses, and more sense of user requests.

Base
Qwen 3.8 27B
Parameter size
27B

DeletedAI

Planned at $30K funding

More efficient reasoning, easier to read responses, more computer use and cyber capabilities, and more sense in designing.

Base
GLM 5.3 Flash
Parameter size
320B

DeletedAI Harness

The agent runtime for DeletedAI models, in development now.

The DeletedAI Harness runs agents on DeletedAI models. Its defining feature is subagent to subagent direct communication: subagents message each other directly instead of routing every exchange through a single orchestrator.

That removes the orchestrator as a bottleneck and a single point of failure, and lets parallel workers coordinate mid-task.

  • Direct subagent messaging

    Subagents talk to each other without an orchestrator in the middle.

  • Shared task context

    Results and state stay visible to every agent working the same task.

  • Built for the model line

    Runs DeletedAI Mini, DeletedAI Flash and the 320B model.

DeletedAI Search Index

Web pages collected, cleaned, and searchable online. Live statistics refresh every few seconds.

The DeletedAI crawler stores page content as Markdown alongside source metadata, deduplicates identical content, and supports search across the collection.

Documents indexed
Loading
Unique content records
Loading
Source hosts
Loading
Collection database
Loading

Pages scanned in the past 24 hours

Loading

Roadmap

Our current experiments and longer-term model plans.

  1. 2026

    DeletedAI Mini

    Iterating and testing small scale with 0.8B parameter models.

  2. 2026

    DeletedAI Search Index

    Web collection and local search for future model retrieval.

  3. 2027

    DeletedAI Harness

    Agent runtime with subagent to subagent direct communication, so parallel agents coordinate without a central orchestrator.

  4. $30K

    DeletedAI Flash

    Planned once funding reaches $30K. Roughly 27B parameters, aimed at more efficient reasoning and responses that are easier to read.

  5. $30K

    DeletedAI

    Planned once funding reaches $30K. 320B parameters, adding computer use, cyber capabilities and stronger design sense on top of the Flash recipe.

Funding

$20 / $30,000

0.07%
$20 of $30,000
  • $5,000
  • $10,000
  • $20,000
  • $30,000