OpenAISpecialist modelLimited access
GPT-Rosalind-5.5
PopularityRank 670 of 875
0out of 100same as last month
Summary
GPT-Rosalind-5.5 is OpenAI’s specialist model for life sciences. It combines agentic tool use with medicinal chemistry, genomics, quantitative biology, and laboratory assistance. The model is multimodal.
Eligible organizations receive global access through OpenAI’s trusted-access structure. The announced variant builds on GPT-5.5. Its safety documentation describes reviewed deployment, not open weights or an unrestricted standard API.
All benchmark results
Every published result with its source, test setup and comparison models.
Show all 14 published benchmark results
Page 1 of 2
Measurements without matching peer values
These measurements have no matching peer values under the same test conditions. Their original values and sources remain available here.
MedChemBench pass@1 (Show measurement, test conditions, and source)
- Source value
- 27.5
- Score
- 27.5
- Metric
- percent
- Unit
- %
- Category
- life-sciences
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- OpenAI system card
- Status
- active
- Retrieved at
- 2026-10-04
- Methodology
- Vendor-reported result. Table 1; multimodal medicinal-chemistry workflow evaluation; GPT-5.5 comparator 25.1%.
GeneBench pass@1 (Show measurement, test conditions, and source)
- Source value
- 21.6
- Score
- 21.6
- Metric
- percent
- Unit
- %
- Category
- life-sciences
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- OpenAI system card
- Status
- active
- Retrieved at
- 2026-10-04
- Methodology
- Vendor-reported result. Table 2; long-horizon genomics and quantitative-biology agentic evaluation; GPT-5.5 comparator 20.4%.
LifeSciBench pass@1 (Show measurement, test conditions, and source)
- Source value
- 63.4
- Score
- 63.4
- Metric
- percent
- Unit
- %
- Category
- life-sciences
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- OpenAI system card
- Status
- active
- Retrieved at
- 2026-10-04
- Methodology
- Vendor-reported result. Table 3; externally expert-judged end-to-end life-sciences research benchmark; GPT-5.5 comparator 58.8%.
- Context
- LifeSciBench and Labwork Bench are distinct benchmarks, not competing values.
Labwork Bench pass@1 (Show measurement, test conditions, and source)
- Source value
- 63.2
- Score
- 63.2
- Metric
- percent
- Unit
- %
- Category
- life-sciences
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- OpenAI system card
- Status
- active
- Retrieved at
- 2026-10-04
- Methodology
- Vendor-reported result. Table 4; proprietary wet-lab protocol perturbation and outcome evaluation; GPT-5.5 comparator 55.8%.
- Context
- LifeSciBench and Labwork Bench are distinct benchmarks, not competing values.
ProtocolQA Open-Ended (Show measurement, test conditions, and source)
- Source value
- 36.4
- Score
- 36.4
- Metric
- pass@1
- Unit
- %
- Benchmark version
- OpenAI GPT-Rosalind-5.5 system card, Table 6
- Category
- life-sciences-capability
- Direction
- Higher is better
- Sample
- 108
- Source type
- vendor-reported
- Evaluator
- OpenAI, GPT-Rosalind-5.5 system card
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. The card converts 108 FutureHouse ProtocolQA multiple-choice questions into open-ended short answers about troubleshooting published laboratory protocols. Exact GPT-Rosalind-5.5 pass@1 result; no confidence interval is published.
- Context
- Provider-reported, task-specific capability result. Higher is better for this task, not evidence of safer behavior or overall model quality. No confidence interval is published.
TroubleshootingBench (Show measurement, test conditions, and source)
- Source value
- 53.31
- Score
- 53.31
- Metric
- pass@1
- Unit
- %
- Benchmark version
- OpenAI GPT-Rosalind-5.5 system card, Table 7
- Category
- life-sciences-capability
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- OpenAI, GPT-Rosalind-5.5 system card
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. OpenAI reports 52 expert-written, non-public biological protocols with three troubleshooting questions per protocol. The report gives the task-construction counts but not the scored denominator; no sample count or confidence interval is inferred.
- Context
- Provider-reported, task-specific capability result on a provider-built dataset. Higher performance on this task does not imply safer behavior or better overall model quality. Its scored denominator and confidence interval are not published.
Biorisk knowledge (Show measurement, test conditions, and source)
- Source value
- 81.1
- Score
- 81.1
- Metric
- cons@32
- Unit
- %
- Benchmark version
- OpenAI GPT-Rosalind-5.5 system card, Table 8
- Category
- bio-risk-capability
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- OpenAI, GPT-Rosalind-5.5 system card
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. The card describes an uncontaminated, unpublished, in-house closed-ended evaluation developed with Gryphon Scientific. It reports cons@32; no scored denominator or confidence interval is published.
- Context
- Biosecurity-relevant capability measurement, not a measure of safety desirability. A higher score means greater capability on this evaluation, not safer behavior or better overall quality. Provider-reported; no N or confidence interval published.
Multi-select Virology Troubleshooting (Show measurement, test conditions, and source)
- Source value
- 51.2
- Score
- 51.2
- Metric
- pass@1
- Unit
- %
- Benchmark version
- OpenAI GPT-Rosalind-5.5 system card, Table 9
- Category
- bio-risk-capability
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- OpenAI, GPT-Rosalind-5.5 system card
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. The card describes fully held-out SecureBio virology troubleshooting questions in a multi-select format. It reports pass@1; the scored denominator and confidence interval are not published.
- Context
- Biosecurity-relevant task capability, not a measure of safety desirability. A higher score does not mean safer behavior or better overall quality. Provider-reported; denominator and confidence interval are not published.
Hard-negative protein binding prediction (Show measurement, test conditions, and source)
- Source value
- 3.13
- Score
- 3.13
- Metric
- pass@4
- Unit
- %
- Benchmark version
- OpenAI GPT-Rosalind-5.5 system card, Table 10
- Category
- bio-risk-capability
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- OpenAI, GPT-Rosalind-5.5 system card
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. OpenAI describes a proprietary non-contaminated hard-negative protein-binding evaluation with five candidates per task, one true binder, 43 protein targets and 492 hotspots, using structures and browser/computer tools. The scored task count is not stated; pass@4 is retained exactly.
- Context
- Biosecurity-relevant protein-design capability, not a measure of safety desirability. A higher pass rate indicates greater capability on this task, not safer behavior or better overall quality. Provider-reported; scored N and confidence interval are not published.
DNA sequence design for transcription-factor binding (Show measurement, test conditions, and source)
- Source value
- 13.64
- Score
- 13.64
- Metric
- pass@1
- Unit
- %
- Benchmark version
- OpenAI GPT-Rosalind-5.5 system card, Table 11
- Category
- bio-risk-capability
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- OpenAI, GPT-Rosalind-5.5 system card
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. The card evaluates DNA sequence design for transcription-factor binding with BPNet and, where available, Basenji2 as oracles, compared with Ledidi. The score is pass@1; the scored denominator and confidence interval are not published.
- Context
- Biosecurity-relevant DNA-design capability, not a measure of safety desirability. A higher pass rate indicates greater capability on this task, not safer behavior or better overall quality. Provider-reported; scored N and confidence interval are not published.
Medicinal chemistry (Show measurement, test conditions, and source)
- Source value
- 27133.5
- Score
- 27133.5
- Metric
- Average output tokens
- Unit
- No unit provided
- Direction
- No direction published
- Source type
- vendor-reported
- Evaluator
- OpenAI GPT-Rosalind-5.5 system card
- Status
- active
- Retrieved at
- 2026-10-19
- Methodology
- Vendor-reported result. Table 1; GPT-Rosalind-5.5 column; row "Average output tokens".
- Context
- This is the source's average output-token count, not a quality score. The table does not state a task-level sample count or confidence interval.
Genomics and quantitative biology (Show measurement, test conditions, and source)
- Source value
- 16721
- Score
- 16721
- Metric
- Average output tokens
- Unit
- No unit provided
- Direction
- No direction published
- Source type
- vendor-reported
- Evaluator
- OpenAI GPT-Rosalind-5.5 system card
- Status
- active
- Retrieved at
- 2026-10-19
- Methodology
- Vendor-reported result. Table 2; GPT-Rosalind-5.5 column; row "Average output tokens".
- Context
- This is the source's average output-token count, not a quality score. The table does not state a task-level sample count or confidence interval.
Page 1 of 2
Popularity compared
How often the model is searched for, used and mentioned, compared with every model of its category.
Popularity index
0 to 100, higher is more popular
20 of 29 models
SAM 3, Meta: 68
EmbeddingGemma 2, Google DeepMind: 63
Unlimited OCR 2, Baidu: 58
Kumo-Tabular, NVIDIA: 55
TeleOCR, TeleAI: 48
Qwen-Drive-1.0-4B, Qwen: 42
AstaBrief-8B, Allen Institute for AI: 37
pplx-embed-v2-context-9b-preview, Perplexity: 35
Agora-2, Odyssey: 33
LFM2.5-VL-3B-DSpark, Liquid AI: 32
Hy-SimulMT-30B-A3B, Tencent Hunyuan: 32
LensVLM-9B, Apple: 31
FLUX 3 Action, Black Forest Labs: 6
Praxis-1, Runway: 0
GPT-5.6 Cyber, OpenAI: 0
Cohere Embed v5 Fast, Cohere: 0
Cohere Embed v5 Pro, Cohere: 0
Cohere Rerank 5 Fast, Cohere: 0
Cohere Rerank 5 Pro, Cohere: 0
GPT-Rosalind-5.5, OpenAI: 0
As of November 2026 · Source: gradually.ai popularity index
Over time
Weekly values from October 18, 2026 to November 5, 2026
GPT-Rosalind-5.5Cohere Embed v5 FastCohere Embed v5 ProCohere Rerank 5 FastCohere Rerank 5 Pro
All specifications
Architecture
Input and output
- Capabilities
- Medicinal chemistry, genomics, quantitative biology, and lab assistance
- Source: Capabilities
Watermarking and provenance
AI watermarking refers to invisible signals embedded in the output. Visible marks and provenance metadata are listed separately. Unknown means there is insufficient evidence for that particular field.
- AI watermarking
- Unknown
- no source
- Watermarking technique
- Not known
- no source
Licence and access
Further details
Sources and data date
Show all 10 sources
Benchmark sources at a glance
Which independent evaluators tested this model. This is coverage, not a quality ranking.
No active result from the reviewed external evaluators can currently be attributed to this model. Other source evidence remains available below.
Without vendor claims and historical or disputed results; those stay labelled in the single values.