$ cat wiki/models/gpt-rosalind.md
GPT-Rosalind
modelupdated 2026-07-21created 2026-05-16
Spec
| Attribute | Value |
|---|---|
| Developer | OpenAI |
| Released | 2026-04-16 (initial; Rosalind Biodefense program launched 2026-05-29) |
| Announced | 2026-04-16 |
| Context window | unknown |
| Pricing | unknown |
| License | unknown |
| Availability | Application-gated sponsored access via Rosalind Biodefense program (vetted developers; U.S. government and allied public-health partners) |
| Tier | Frontier reasoning model (domain-specialized) |
| Domain | Life sciences — biology, drug discovery, translational medicine |
| Named after | Rosalind Franklin (DNA structure research) |
Release Date
2026-04-16 (model release); 2026-05-29 (Biodefense program launch)
Capabilities (per OpenAI)
- Stronger reasoning about molecules, proteins, genes, and disease biology vs. general GPT models
- Optimized for scientific workflows: improved tool use + deeper understanding across chemistry, protein engineering, genomics
- Tailored for research across biology, drug discovery, and translational medicine
Benchmarks
Specific benchmark numbers not publicly disclosed by OpenAI. Benchmarked internally against GPT-5.x on life sciences tasks — described as meaningfully outperforming general models. (source)
Use Cases
General Life Sciences (since April 2026)
- Drug discovery & translational medicine
- Protein engineering
- Genomics research
- Chemistry reasoning
Rosalind Biodefense Program (since May 29, 2026)
- Epidemiological modeling
- Biosurveillance & early detection
- Non-pharmaceutical interventions (NPIs)
- Medical countermeasure development
- Pandemic preparedness planning
- Biothreat characterization
Update: New Capabilities (2026-06-04)
OpenAI upgraded GPT-Rosalind with GPT-5.5 as the base model and added two new plugins (source):
- Base model upgrade: GPT-5.5 (released April 24, 2026) — stronger agentic coding and tool-use
- Life Sciences Research Plugin: sourced evidence retrieval + biological interpretation in one workspace
- Life Sciences NGS Analysis Plugin: bioinformatics execution for Next-Generation Sequencing data
- Biodefense access expansion: select U.S. government and allied public-health partners added
Performance gains (OpenAI internal): broad improvements on biology expert tasks, medicinal chemistry, quantitative biology, wet lab protocol assistance (improved token efficiency vs. GPT-5.5 baseline).
Rosalind Biodefense Program
OpenAI launched Rosalind Biodefense on 2026-05-29 — a program providing sponsored, application-gated access to GPT-Rosalind for vetted developers and government partners (source):
- Access model: Application-based; OpenAI sponsors access + provides launch support
- Government engagement: White House and federal agencies briefed; public-health-focused agencies being added
- Launch partners:
- Lawrence Livermore National Laboratory — biodefense applications
- Johns Hopkins Applied Physics Laboratory — protein-engineering platform for therapeutic screening + biothreat characterization
- CEPI — vaccine initiative, pandemic preparedness
- Strategic framing: Three pillars — (1) back biodefense startups, (2) engage outside stakeholders, (3) develop internal biological risk evaluation systems
Compared To
- DeepMind AlphaFold / Gemini for Science — Google's parallel science AI push (different modality: protein structure vs. general life sciences reasoning)
- Anthropic Gates Foundation partnership (2026-05-14) — $200M for global health with Claude → both labs betting on AI for public health, different approaches
- Claude Opus 4.7 — same day release (2026-04-16), general-purpose vs. domain-specialized contrast
Significance
- "Year of Science" takes concrete form: The core asset of the 2026 slogan is connected to government / public-health infrastructure
- Government AI partnership pattern: First case of institutionalizing an AI model as public-health infrastructure (alongside the DOE collaboration)
- Domain-specialized frontier model line: Maintaining a naming and access structure separate from general GPT-5.x → a signal of building a dedicated model portfolio for the science domain
Open Questions
- Quantitative benchmark comparison against general GPT-5.x not disclosed — to be supplemented later
- Plans to expand access to non-U.S. governments / research institutions?
- Plans to expand into other science domains such as physics and chemistry?