AI for Bio · Model

Frontier general-purpose LLMs

Category entry for general frontier LLMs (Claude, GPT, Gemini) tested on bio benchmarks

Text, LLMs & agentsBig-tech labUnited States, United KingdomActive中文

Overview

NameFrontier general-purpose LLMs
Full nameNo separate expanded name
Ambiguous names (not auto-linked)Claude, GPT, Gemini
FamilyCategory entry for general closed LLMs from Anthropic, OpenAI and Google as test subjects; not one model, no single first release
  • Claude 3.5 Sonnet · †
  • GPT-4o · †
  • Claude Fable 5.2 / Opus 5.5 / Sonnet 5.5 / Haiku 5.5 · No sourced date†
OrganizationAnthropic, OpenAI, Google DeepMind
Organization typeBig-tech lab
CountryUnited States, United Kingdom
Model classGeneral-purpose LLM
DomainText, LLMs & agents
First releaseNo sourced date† · new version
Basis: No single first release for a family entry; earliest cited members c. 2024-05/06 (unverified)
Other milestonesNo other milestone recorded.
Latest updateNo sourced date† · successor
Basis: Current models listed on the developer model page
Developer page lists Fable 5.2, Opus 5.5, Sonnet 5.5 and Haiku 5.5 as current, with no release dates; OpenAI and Google dates unverified
Note: The earlier 2026-11-09 was back-derived from a retirement floor, not a release date; the four floors differ, so not one release day source
The most recent sourced event, not a claim that nothing changed since.
StatusActive · Vendors keep releasing new models
OpennessCode: Closed
Weights: Closed
Proprietary models accessed through APIs
Benchmarks reported5 counted (board 2, self 2, retrospective 1, unconfirmed 1)
Wet-lab validationNone recorded
Unverified fields† first_release, latest_update, family.members, benchmarks_reported[2], benchmarks_reported[5]
As of2026-11-16

What it does

This site treats general frontier LLMs such as Claude, GPT and Gemini as one category entry, recording how they fare on bio benchmarks such as LAB-Bench and BixBench. They are test subjects, not biology-specific models.

Benchmarks reported

Every row is one pairing from the survey's relation file; a score belongs to the benchmark page's ranking, where it is shown with metric, eval set, source type and date. 5 of 6 pairings are counted.

BenchmarkRelationEval set / protocolSource typeNote and evidence
18 LAB-Bench / LABBench2Official leaderboard or resultsLABBench2BoardLABBench2 official board entries; family entry. · evidence
29 BixBenchOfficial leaderboard or resultsBixBench3BoardSuccessor's official board; counted as one family cell · evidence
14 PubMedQADeveloper report, official protocol—Self-reportedGemini 3 figures self-reported in the MedGemma 1.5 report; earlier board entries (GPT-4, Claude 3) were self-submitted. †partly verified · evidence
42 BioMysteryBenchDeveloper report, official protocol—Self-reportedThe publisher runs the benchmark itself and reports scores; the cross-vendor board is run by a third party. · evidence
02 ProteinGymRetrospective dataset use—Third-partyScores from a second-hand aggregator; the entry is a family, not one model †unverified · no source link
23 BioASQUnconfirmed (not counted)—Unconfirmed sourceThe overview ranks submitted systems, not the underlying LLMs. †disputed · evidence

Other evaluations (outside the 44)

None recorded.

Developers report: Source states no developer claim

Independent evidence: No independent evidence found. No independent evidence found

Training data

Group in Figure 7: Not disclosed.

ClassValueDetail
a. Public databasesNot disclosed
b. Private or licensed dataNot disclosed
c. Own wet-lab data used for trainingNo
d. Wet lab used only for validationNo

Wet-lab validation

None. The developers report no wet-lab validation.

Background

Reading notes

  1. LAB-Bench and BixBench come from Edison, itself a competitor
  2. Scores shift with thinking and tool settings; not comparable across sources

Caveats

  1. Not a single model: first release and latest update only make sense per member
  2. OpenAI and Google model names and dates not verified from primary pages

Sources

  1. Official page platform.claude.com (company blog)
  2. Official page advances.edisonscientific.com (company blog)
  3. arXiv preprint 2604.09554 (preprint)
  4. Anthropic BioMysteryBench article (company blog)

Corrections

1 correction
FieldWasNowBasisFound
latest_update.descriptionClaude Fable 5.2 released "alongside" Opus 5.5 etc., dated 2026-11-09No release date on the developer page; 2026-11-09 was back-derived from a retirement floor and the four models were not released on one daysource2026-11-16

Part of the AI for Bio benchmarks survey · all 112 models · open in the explorer. Record as of 2026-11-16.