Fractal logo

Fractal models

Fractal is a listed enterprise AI company whose AI Research team builds the open Fathom reasoning models and the Vaidya healthcare reasoning models.

New York, United StatesFounded 2000fractal.ai

Releases
3
Open weights
2
Latest
Feb 19, 2026

February 2026

1 release
  1. ReasoningProprietary

    Vaidya 2.0

    Vaidya 2.0 is Fractal's second generation of healthcare reasoning models, post-trained for clinical and citizen-facing health workflows and offered through Vaidya.ai.

    Modalities
    text → text
    At launch
    Consumer app
    • Fractal reported a score of 50.1 on OpenAI's HealthBench (hard), above GPT-5 and Google's Gemini 3 Pro in its comparison
    • Targets uses such as emergency assist, symptom checking, patient journey support and doctor assist
    • Unveiled at the India AI Impact Summit 2026 in New Delhi
    • By November 2026 vaidya.ai also listed a paid developer API serving the model as Vaidya-v2
    Source: Fractal launches Vaidya 2.0, outperforming leading frontier models on Healthcare AI Benchmarks - Fractal Analytics

September 2025

1 release
  1. ReasoningOpen weights

    Fathom-DeepResearch (Fathom-Search-4B and Fathom-Synthesizer-4B)

    Fathom-DeepResearch is Fractal's open agentic research system built from two 4B models fine-tuned from Qwen3-4B: Fathom-Search-4B for long-horizon live web search and Fathom-Synthesizer-4B for writing research reports.

    Modalities
    text → text
    Parameters
    Two 4B models
    License
    MIT
    At launch
    Open weights
    • Fractal reported state-of-the-art open-weights results on SimpleQA, FRAMES, WebWalkerQA and Seal0
    • Fractal also reported beating several closed DeepResearch agents on DeepResearch-Bench
    • Released with the DuetQA and DeepResearch-SFT datasets and a technical report
    • The Fathom-Search-4B paper was accepted to the SEA workshop at NeurIPS 2025
    Source: FractalAIResearch/Fathom-Search-4B · Hugging Face

May 2025

1 release
  1. ReasoningOpen weights

    Fathom-R1-14B

    Fathom-R1-14B is Fractal's open 14B math reasoning model, post-trained from DeepSeek-R1-Distill-Qwen-14B to perform well within a 16K-token reasoning budget.

    Modalities
    text → text
    Parameters
    14B
    License
    MIT
    At launch
    Open weights
    • Fractal reported 52.71% pass@1 on AIME 2025 and 35.26% on HMMT 2025
    • Post-training cost about $499 using supervised fine-tuning and model merging, per Fractal
    • Released with a sibling, Fathom-R1-14B-RS, trained with RL and SFT for about $967, plus datasets and recipes
    • Evaluated under a 16,384-token output budget
    Source: FractalAIResearch/Fathom-R1-14B · Hugging Face

Cite and reuse this data

These release records are free to reuse under CC BY 4.0, including commercially. Credit The AI Engineering and link to this page's URL.

Dates are the labs' official announcement dates, and every record links to its source. List fields in the CSV are separated by " | ".

Citation

The AI Engineering Editors. "Fractal models: release history and timeline." The AI Engineering, updated Feb 19, 2026. https://theaiengineering.com/releases/fractal

HTML link

<a href="https://theaiengineering.com/releases/fractal">Fractal models: release history and timeline</a> (The AI Engineering)
Full dataset
1,250 releases from 73 labs
Coverage
to
Updated
License
CC BY 4.0
Publisher
The AI Engineering

Other labs