Subscribe

Fx-Bio-1031

The Endless Frontier released the full weights of Fx-Bio-1031, a 180B-parameter agentic bioinformatics model post-trained from Qwen3.8-Flash-Next, scoring 70.59% Pass@5 on BioMysteryBench Difficult.

Fx-Bio-1031 is post-trained from Alibaba's Qwen3.8-Flash-Next to plan and run multi-step bioinformatics analyses on its own, including writing scripts, installing software, querying NCBI and Ensembl, and checking its results. The company says it used a recursive self-improvement pipeline called RSI-OS across evaluation, training data and tool environments. It targets hospitals, labs and drug developers that want to run analyses on local hardware.

Date
Friday 13 November 2026
Lab
The Endless Frontier
Kind
open-weights
Access
open weights (restricted license)

Figures

MeasureValueMeasured by
BioMysteryBench Difficult subset, Pass@570.59%
Opus-5 52.94%, GPT-6-Astra 47.06%
company
BioMysteryBench expert-solvable subset93.15%
Opus-5 94.52%
company
BiomniBench, Pass@389.50%
Opus-5 87.42%
company
BixBench345.76
GPT-6-Astra 48.70, Opus-5 40.60
company
Total language parameters180B
125B MoE backbone with 6B active per token, 51B n-gram embedding, 4B multi-token prediction module
company
Context length524,288 tokens
released configuration
company

Weights are about 360GB in BF16 under the Qwen Community License 1.0; benchmark results come from the company's own evaluation framework.

Sources

  1. pandaily.com/endless-frontier-fx-bio-1031-open-weights-bioinformatics-agent-model-biomyste

This record was checked against its sources on 14 November 2026. How we check

Read the daily brief for 13 November 2026