Virology Capabilities Test (VCT)
Grok 4.6 67.4% leads since Aug 2026 Only in system cards
Models answer 322 multimodal virology troubleshooting questions written by expert virologists. Multiple-select variant requires all correct options.
Anthropic cards call it Multimodal virology and use the multiple-select variant.
In the authors’ words
We present the Virology Capabilities Test (VCT), a large language model (LLM) benchmark that measures the capability to troubleshoot complex virology laboratory protocols. Constructed from the inputs of dozens of PhD-level expert virologists, VCT consists of 322 multimodal questions covering fundamental, tacit, and visual knowledge that is essential for practical work in virology laboratories. VCT is difficult: expert virologists with access to the internet score an average of 22.1% on questions specifically in their sub-areas of expertise. However, the most performant LLM, OpenAI's o3, reaches 43.8% accuracy, outperforming 94% of expert virologists even within their sub-areas of specialization. The ability to provide expert-level virology troubleshooting is inherently dual-use: it is useful for beneficial research, but it can also be misused. Therefore, the fact that publicly available models outperform virologists on VCT raises pressing governance considerations. We propose that the capability of LLMs to provide expert-level troubleshooting of dual-use virology work should be integrated into existing frameworks for handling dual-use technologies in the life sciences.
Made by SecureBio and Center for AI Safety Released Apr 2025 Hazardous knowledge Public tasks plus a private test set Last in a system card Oct 2026
Last model added Oct 2026
Scores refreshed 2 Nov 2026
- Level
- Expert Specialists with years of training, often over hours of work
- Scored
- Checked automatically A program checks the result against an answer key, unit tests or the expected final state.
- Answers
- Multiple choice
28
Models scored in lab cards
32
System cards report it, from 4 labs
Grok 4.6 leads with 67.4%
Best score per model as labs report them in their own cards. Labs pick their own settings, so these compare less cleanly than independent runs.
- 67.4 Grok 4.6
- 65.5 Grok 4.5
- 63.0 Grok 4.7
- 61.0 Grok 4.1
- 60.0 Grok 4
- 59.0 Claude Opus 5
- 59.0 Claude Opus 5.5
- 58.0 Claude Sonnet 5.5
- 57.8 GPT-6 Astra
- 57.4 Claude Mythos Preview
- 55.9
Muse Wonder Full - 54.5 Grok 4 Fast
- 53.5 GPT-5.6 Sol
- 52.0 Muse Spark 1.1
- 51.7 Muse Spark 1.3
- 50.0 Claude Opus 4.7
- 49.7 Muse Spark
- 48.3 Claude Opus 4.6
- 48.0 Claude Haiku 5.5
- 47.7 Claude Opus 4.5
- 47.0 Claude Opus 4.8
- 45.0 Claude Sonnet 5
- 41.4 GPT-5
- 39.7 Claude Sonnet 4.5
- 39.7 Claude Sonnet 4.6
- 32.0 Claude Haiku 4.5
- 28.7 grok-code-fast-1
- 23.8 Code World Model (CWM)
- 1 Grok 4.6 67.4%
- 2 Grok 4.5 65.5%
- 3 Grok 4.7 63%
- 4 Grok 4.1 61%
- 5 Grok 4 60%
- 6 Claude Opus 5 59%
- 7 Claude Opus 5.5 59%
- 8 Claude Sonnet 5.5 58%
- 9 GPT-6 Astra 57.8%
- 10 Claude Mythos Preview 57.4%
- 11 Muse Wonder Full 55.9%
- 12 Grok 4 Fast 54.5%
- 13 GPT-5.6 Sol 53.5%
- 14 Muse Spark 1.1 52%
- 15 Muse Spark 1.3 51.7%
- 16 Claude Opus 4.7 50%
- 17 Muse Spark 49.7%
- 18 Claude Opus 4.6 48.3%
- 19 Claude Haiku 5.5 48%
- 20 Claude Opus 4.5 47.7%
- 21 Claude Opus 4.8 47%
- 22 Claude Sonnet 5 45%
- 23 GPT-5 41.4%
- 24 Claude Sonnet 4.5 39.7%
- 25 Claude Sonnet 4.6 39.7%
- 26 Claude Haiku 4.5 32%
- 27 grok-code-fast-1 28.7%
- 28 Code World Model (CWM) 23.8%
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Grok 4.6 | 67.4% | Aug 2026 | Model card | |
| 2 | Grok 4.5 | 65.5% | Jul 2026 | Model card | |
| 3 | Grok 4.7 | 63% | Sep 2026 | Model card | |
| 4 | Grok 4.1 | 61% | Nov 2025 | Model card | |
| 5 | Grok 4 | 60% | Aug 2025 | Model card | |
| 6 | Claude Opus 5 | 59% | Jul 2026 | Model card | |
| 7 | Claude Opus 5.5 | 59% | Sep 2026 | Model card | |
| 8 | Claude Sonnet 5.5 | 58% | Sep 2026 | Model card | |
| 9 | GPT-6 Astra | 57.8% | Sep 2026 | Model card | |
| 10 | Claude Mythos Preview | 57.4% | Apr 2026 | Model card | |
| 11 | Muse Wonder Full | 55.9% | Oct 2026 | Model card | |
| 12 | Grok 4 Fast | 54.5% | Sep 2025 | Model card | |
| 13 | GPT-5.6 Sol | 53.5% | Jun 2026 | Model card | |
| 14 | Muse Spark 1.1 | 52% | Jul 2026 | Model card | |
| 15 | Muse Spark 1.3 | 51.7% | Oct 2026 | Model card | |
| 16 | Claude Opus 4.7 | 50% | Apr 2026 | Model card | |
| 17 | Muse Spark | 49.7% | Jun 2026 | Model card | |
| 18 | Claude Opus 4.6 | 48.3% | Feb 2026 | Model card | |
| 19 | Claude Haiku 5.5 | 48% | Oct 2026 | Model card | |
| 20 | Claude Opus 4.5 | 47.7% | Nov 2025 | Model card | |
| 21 | Claude Opus 4.8 | 47% | May 2026 | Model card | |
| 22 | Claude Sonnet 5 | 45% | Jun 2026 | Model card | |
| 23 | GPT-5 | 41.4% | Aug 2025 | Model card | |
| 24 | Claude Sonnet 4.5 | 39.7% | Sep 2025 | Model card | |
| 25 | Claude Sonnet 4.6 | 39.7% | Feb 2026 | Model card | |
| 26 | Claude Haiku 4.5 | 32% | Oct 2025 | Model card | |
| 27 | grok-code-fast-1 | 28.7% | Aug 2025 | Model card | |
| 28 | Code World Model (CWM) | 23.8% | Oct 2025 | Model card |
The benchmark's own leaderboard has 7 models. We plot every source in the charts below.
2 models have held number one since its release in Apr 2025
Lab claims count from the day the card came out, independent runs from the model’s release. We draw one row each for the benchmark’s own leaderboard and what labs claim in their cards.
Google DeepMind OpenAI xAI
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
On the benchmark’s own leaderboard Reported by the lab in its model card
Best so far
Labs report it only in system cards
We read the headline results of 479 launch posts from 51 labs, and none of them list this benchmark. Some system cards do, below.
32 system cards report it, from 4 labs
Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.
| Date | Document | Score | As the document writes it | Section |
|---|---|---|---|---|
| 23 Oct 2026 | Muse Spark 1.3 system card | 51.7 | VCT accuracy, Muse Spark 1.3 (API) max, multiple-response MC (Table 1, Figure 1) | Bio and chem risk |
| 12 Oct 2026 | Muse Wonder Full model card | 55.9 | SecureBio VCT (CBRN Capability Evaluations table, section 4) | Bio and chem risk |
| 7 Oct 2026 | Claude Haiku 5.5 system card | 48 | Multimodal virology (VCT), accuracy 0.48 on the 350-question set, printed on chart (Figure 2.2.2.A) | Bio and chem risk |
| 28 Sep 2026 | Claude Sonnet 5.5 system card | 58 | Multimodal virology (VCT), accuracy 0.58 on the 350-question set (2.2.2, Figure 2.2.2.A) | Bio and chem risk |
| 22 Sep 2026 | Claude Opus 5.5 system card | 59 | Multimodal virology (VCT), accuracy 0.59 on the 350-question set, highest-scoring snapshot (2.2.3.1, Figure 2.2.3.1.A) | Bio and chem risk |
| 21 Sep 2026 | Grok 4.7 model card | 63 | Virology Capabilities Test (VCT), accuracy, high, without safeguards (7.2) | Bio and chem risk |
| 3 Sep 2026 | GPT-6 Astra system card | 57.8 | SecureBio VCT, refusal-corrected accuracy, pre-release checkpoint (10.1.1.4) | Bio and chem risk |
| 12 Aug 2026 | Grok 4.6 model card | 67.4 | Virology Capabilities Test (VCT), accuracy, high, without safeguards (8.1) | Bio and chem risk |
| 24 Jul 2026 | Claude Opus 5 system card | 59 | Multimodal virology (VCT), score 0.59, highest-scoring snapshot (2.2.4, Figure 2.2.4.A) | Bio and chem risk |
| 14 Jul 2026 | Grok 4.5 model card | 65.5 | Virology Capabilities Test (VCT), accuracy, high, without safeguards (8.1) | Bio and chem risk |
| 9 Jul 2026 | GPT-5.6 Sol system card | 53.5 | SecureBio Virology Capabilities Test, strongest reported configuration of Sol or its railfree version (9.1.1.8) | Bio and chem risk |
| 9 Jul 2026 | Muse Spark 1.1 system card | 52 | VCT, Muse Spark 1.1 Helpful (Table 1, Figure 1) | Bio and chem risk |
| 30 Jun 2026 | Claude Sonnet 5 system card | 45 | Multimodal virology (VCT), score 0.45, expert baseline 0.221 (2.2.4) | Bio and chem risk |
| 25 Jun 2026 | GPT-5.6 Sol system card | 53.5 | SecureBio Virology Capabilities Test, strongest reported configuration of Sol or its railfree version (9.1.1.8) | Bio and chem risk |
| 12 Jun 2026 | Muse Spark system card | 49.7 | VCT (Table 1, Figure 1) | Bio and chem risk |
| 28 May 2026 | Claude Opus 4.8 system card | 47 | Multimodal virology (VCT), Virology Capabilities Test score 0.470 (Table 2.2.3.B, 2.2.4) | Bio and chem risk |
| 16 Apr 2026 | Claude Opus 4.7 system card | 50 | Multimodal virology (VCT), score 0.50, highest score across snapshots, printed on chart (2.2.5.2, Figure 2.2.5.2.A) | Bio and chem risk |
| 7 Apr 2026 | Claude Mythos Preview system card | 57.4 | Multimodal virology (VCT), score 0.574 (2.2.5.4, Figure 2.2.5.4.A) | Bio and chem risk |
| 17 Feb 2026 | Claude Sonnet 4.6 system card | 39.7 | Multimodal virology (VCT), multiple select variant, mean score 0.397 (6.2.2.2) | Bio and chem risk |
| 5 Feb 2026 | Claude Opus 4.6 system card | 48.3 | Multimodal virology (VCT), multiple select variant, mean score 0.483 (8.2.4.2) | Bio and chem risk |
| 24 Nov 2025 | Claude Opus 4.5 system card | 47.7 | Multimodal virology (VCT), multiple select variant, mean score 0.4771 (7.2.4.2) | Bio and chem risk |
| 17 Nov 2025 | Grok 4.1 model card | 61 | VCT, accuracy 0.61, text-only questions, Grok 4.1 Thinking, safeguards removed (Table 4) | Bio and chem risk |
| 15 Oct 2025 | Claude Haiku 4.5 system card | 32 | VMQA/VCT, score 0.32 (6.2.3.1) | Bio and chem risk |
| 8 Oct 2025 | Code World Model (CWM) technical report | 23.8 | VCT, text-only subset of 101 questions (Table 13) | Bio and chem risk |
| 29 Sep 2025 | Claude Sonnet 4.5 system card | 39.7 | Multimodal virology (VMQA/VCT), multiple select, 10-shot, 39.7% (9.2.4.2, Figure 9.2.4.2.A 0.397) | Bio and chem risk |
| 19 Sep 2025 | Grok 4 Fast model card | 54.5 | VCT, accuracy 54.5, text-only portion (Table 3) | Bio and chem risk |
| 26 Aug 2025 | grok-code-fast-1 model card | 28.7 | VCT, accuracy 28.7 (Table 3) | Bio and chem risk |
| 20 Aug 2025 | Grok 4 model card | 60 | VCT, accuracy 0.60, text-only portion, Grok 4 API (Table 3) | Bio and chem risk |
| 7 Aug 2025 | GPT-5 system card | 41.4 | SecureBio Virology Capabilities Test, gpt-5-thinking July 13, mean of n=10 (Table 15) | Bio and chem risk |
| 22 May 2025 | Claude Sonnet 4 system card | Multimodal virology VMQA/VCT, multiple select, 0, 5 and 10 shots, extended thinking, chart without printed values (Figure 7.2.4.4.A) | Bio and chem risk | |
| 22 May 2025 | Claude Opus 4 system card | Multimodal virology VMQA/VCT, multiple select, 0, 5 and 10 shots, extended thinking, chart without printed values (Figure 7.2.4.4.A) | Bio and chem risk | |
| 24 Feb 2025 | Claude 3.7 Sonnet system card | Multimodal virology (VCT), multiple select, 0 and 5 shots with Bo1 to Bo10, 32k thinking, chart without printed values. Text says above the human baseline (Figure 16) | Bio and chem risk |
Versions
More in Biology and chemistry
- SUPERChem In use Models solve 500 expert-written chemistry reasoning problems, offered in multimodal and text-only forms. Expert solution paths also allow a reasoning path score.
Tencent Hy4 preview 66.4% - GeneBench Pro In use Agents analyze messy simulated genomics and biology datasets to estimate a decision-relevant quantity across 129 problems. Scored automatically against known answers. GPT-6 Astra 37.1%
- LifeSciBench In use Models complete 750 expert-written life science research tasks, often with attached figures, tables, or sequence files. Graded against expert rubrics as score and pass rate. GPT-6 Astra 60.3%
- MolecularIQ In use Models answer questions about molecular graphs, such as counting, indexing and constrained generation, and a symbolic checker verifies each answer.
Intern-S2-397B 62.4% - TOMG-Bench In use Models generate molecules from natural language requests in three tasks: molecule editing, molecule optimization and customized generation. Each output is checked for validity and for meeting the requested properties. Intern-S2-397B 66.8%
- BABE Fading Models reason from experimental results in peer-reviewed biology papers to draw conclusions, with causal and cross-scale inference.
Seed2.0 Lite (0428) 57.9%