GPT-56T 861 —
MUSE-SPK 837 —
GPT-56SC 789 -0.1%
GLM-5 781 —
CL-OP55X 779 -0.1%
GROK-46H 779 -0.1%
QWEN-38X 748 —
GPT-6A 743 —
KIMI-K3X 742 —
CL-FAB5H 697 -0.1%
CL-OP5H 674 -0.1%
GEM-38FH 672 —
CL-OP5X 669 -0.1%
CL-OP55H 667 -0.1%
CL-OP46H 656 -0.2%
CL-OP47H 647 -0.2%
GPT-56S 617 -0.2%
GEM-37FH 609 -0.2%
GEM-36FH 592 -0.2%
CL-OP48H 587 -0.2%
CL-OP47 580 -0.2%
GEM-35FH 579 -0.2%
GPT-55H 540 -0.2%
INKL 531 —
GEM-31P 511 -0.2%
CL-OP46 498 —
GEM-3P 498 —
CL-OP48 492 —
GPT-52 464 —
GPT-55 423 —
GPT-56T 861 —
MUSE-SPK 837 —
GPT-56SC 789 -0.1%
GLM-5 781 —
CL-OP55X 779 -0.1%
GROK-46H 779 -0.1%
QWEN-38X 748 —
GPT-6A 743 —
KIMI-K3X 742 —
CL-FAB5H 697 -0.1%
CL-OP5H 674 -0.1%
GEM-38FH 672 —
CL-OP5X 669 -0.1%
CL-OP55H 667 -0.1%
CL-OP46H 656 -0.2%
CL-OP47H 647 -0.2%
GPT-56S 617 -0.2%
GEM-37FH 609 -0.2%
GEM-36FH 592 -0.2%
CL-OP48H 587 -0.2%
CL-OP47 580 -0.2%
GEM-35FH 579 -0.2%
GPT-55H 540 -0.2%
INKL 531 —
GEM-31P 511 -0.2%
CL-OP46 498 —
GEM-3P 498 —
CL-OP48 492 —
GPT-52 464 —
GPT-55 423 —
← Back to feed

OpenAI's GPT-Rosalind Posts 0.751 on BixBench — First Domain-Specific Model for Drug Discovery, Restricted to US Enterprise

OpenAI announced GPT-Rosalind on April 16, its first model built specifically for the life sciences. Named after Rosalind Franklin — the British chemist whose X-ray crystallography work helped establish the structure of DNA — the model is a frontier reasoning system fine-tuned for biochemistry, protein engineering, genomics, and translational medicine research. It is currently available only to qualified US enterprise customers through a Trusted Access programme.

The move follows GPT-5.4-Cyber, OpenAI’s April defensive cybersecurity model, establishing a pattern: domain-specific fine-tunes of frontier reasoning infrastructure, distributed through controlled access pipelines rather than public APIs.

Benchmark Results

On BixBench — a bioinformatics benchmark covering real-world data analysis tasks — GPT-Rosalind scores 0.751 on Pass@1. GPT-5.4 scores 0.732 on the same benchmark. Gemini 3.1 Pro scores 0.550. On LABBench2, which tests literature review and protocol design across 11 scientific research tasks, Rosalind outperforms GPT-5.4 on 6 of the 11 tasks, with the largest margin on CloningQA — end-to-end molecular cloning protocol design.

In a collaboration with Dyno Therapeutics using unpublished RNA sequences, the model’s top-10 predictions ranked in the 95th percentile relative to human expert baselines on sequence prediction tasks. OpenAI cautions that all outputs should be treated as preliminary and independently validated.

BenchmarkGPT-RosalindGPT-5.4Gemini 3.1 Pro
BixBench Pass@10.7510.7320.550
LABBench2 tasks won6 / 11——

What It Does

The model is designed for multi-step research workflows: querying scientific databases, parsing literature, generating hypotheses, planning experiments, and suggesting downstream validation steps. It connects to more than 50 scientific tools and data sources through a new Life Sciences research plugin for Codex — a plugin that is free and works with standard API models, not only Rosalind.

OpenAI describes it as tuned for skepticism. The model is designed to flag uncertainty rather than confabulate experimental pathways, a meaningful design constraint given the stakes in drug discovery workflows. Closed weights, no reasoning trace disclosure, and role-based access controls enforce this positioning.

Launch partners include Amgen, Moderna, the Allen Institute for Brain Science, and Thermo Fisher Scientific.

Competitive Context

OpenAI is not entering an empty space. Google DeepMind’s AlphaFold and Isomorphic Labs’ IsoDDE dominate molecular structure prediction. Anthropic has a Claude for Life Sciences offering. The strategic difference: OpenAI is positioning Rosalind as a general-purpose scientific reasoning engine that operates across biochemistry, genomics, and translational medicine simultaneously, rather than a narrow prediction engine for one modality.

The restricted rollout is both a biosecurity decision and a product strategy. A model that reasons about pathogens and protein engineering carries dual-use risk. Limiting it to vetted US customers addresses regulators; charging enterprise rates addresses the economics of fine-tuning and managed access infrastructure. Whether the model extends to academic research institutions or international markets is not yet addressed.

GPT-Rosalind is described as the first in a series. OpenAI says future versions will increase support for long-duration, tool-intensive research workflows — the same trajectory as its general-purpose models but pointed at laboratory automation and clinical trial acceleration.