DiscoveryWorld

Virtual interactive environment where agents formulate hypotheses, design controlled experiments, gather data, and discover scientific laws.27 Oct 2026

One row per model and compatible score measure. Mean = equal-weight arithmetic average of imported reports. Distinct configurations reported for the same model in a source table remain separate. Reports can overlap; the range is not a confidence interval. Unbundle to inspect each observation.
		Evidence	What the result shows
GPT-4o / ReACT →OpenAI	10% 1 report	DiscoveryWorld · Average (Challenge) · Explanatory knowledge[1]DiscoveryWorld · GPT-4o agent scaffolds	Leads this displayed comparison. Source notes Source cell: 0.10
GPT-4o / Hypothesizer →OpenAI	8% 1 report	DiscoveryWorld · Average (Challenge) · Explanatory knowledge[1]DiscoveryWorld · GPT-4o agent scaffolds	#2 of 3 in this displayed comparison; 2 pp below the leader. Source notes Source cell: 0.08
GPT-4o / Plan+Execute →OpenAI	1% 1 report	DiscoveryWorld · Average (Challenge) · Explanatory knowledge[1]DiscoveryWorld · GPT-4o agent scaffolds	#3 of 3 in this displayed comparison; 9 pp below the leader. Source notes Source cell: 0.01
GPT-4o / ReACT →OpenAI	63% 1 report	DiscoveryWorld · Average (Challenge) · Procedure[1]DiscoveryWorld · GPT-4o agent scaffolds	Leads this displayed comparison. Source notes Source cell: 0.63
GPT-4o / Plan+Execute →OpenAI	50% 1 report	DiscoveryWorld · Average (Challenge) · Procedure[1]DiscoveryWorld · GPT-4o agent scaffolds	#2 of 3 in this displayed comparison; 13 pp below the leader. Source notes Source cell: 0.50
GPT-4o / Hypothesizer →OpenAI	49% 1 report	DiscoveryWorld · Average (Challenge) · Procedure[1]DiscoveryWorld · GPT-4o agent scaffolds	#3 of 3 in this displayed comparison; 14 pp below the leader. Source notes Source cell: 0.49
GPT-4o / ReACT →OpenAI	18% 1 report	DiscoveryWorld · Average (Challenge) · Task completion[1]DiscoveryWorld · GPT-4o agent scaffolds	Leads this displayed comparison. Source notes Source cell: 0.18
GPT-4o / Plan+Execute →OpenAI	15% 1 report	DiscoveryWorld · Average (Challenge) · Task completion[1]DiscoveryWorld · GPT-4o agent scaffolds	#2 of 3 in this displayed comparison; 3 pp below the leader. Source notes Source cell: 0.15
GPT-4o / Hypothesizer →OpenAI	8% 1 report	DiscoveryWorld · Average (Challenge) · Task completion[1]DiscoveryWorld · GPT-4o agent scaffolds	#3 of 3 in this displayed comparison; 10 pp below the leader. Source notes Source cell: 0.08
GPT-4o / Hypothesizer →OpenAI	34% 1 report	DiscoveryWorld · Average (Easy) · Explanatory knowledge[1]DiscoveryWorld · GPT-4o agent scaffolds	Leads this displayed comparison. Source notes Source cell: 0.34
GPT-4o / ReACT →OpenAI	25% 1 report	DiscoveryWorld · Average (Easy) · Explanatory knowledge[1]DiscoveryWorld · GPT-4o agent scaffolds	#2 of 3 in this displayed comparison; 9 pp below the leader. Source notes Source cell: 0.25
GPT-4o / Plan+Execute →OpenAI	11% 1 report	DiscoveryWorld · Average (Easy) · Explanatory knowledge[1]DiscoveryWorld · GPT-4o agent scaffolds	#3 of 3 in this displayed comparison; 23 pp below the leader. Source notes Source cell: 0.11
GPT-4o / ReACT →OpenAI	59% 1 report	DiscoveryWorld · Average (Easy) · Procedure[1]DiscoveryWorld · GPT-4o agent scaffolds	Leads this displayed comparison. Source notes Source cell: 0.59
GPT-4o / Hypothesizer →OpenAI	56% 1 report	DiscoveryWorld · Average (Easy) · Procedure[1]DiscoveryWorld · GPT-4o agent scaffolds	#2 of 3 in this displayed comparison; 3 pp below the leader. Source notes Source cell: 0.56
GPT-4o / Plan+Execute →OpenAI	56% 1 report	DiscoveryWorld · Average (Easy) · Procedure[1]DiscoveryWorld · GPT-4o agent scaffolds	#2 of 3 in this displayed comparison; 3 pp below the leader. Source notes Source cell: 0.56
GPT-4o / ReACT →OpenAI	38% 1 report	DiscoveryWorld · Average (Easy) · Task completion[1]DiscoveryWorld · GPT-4o agent scaffolds	Leads this displayed comparison. Source notes Source cell: 0.38
GPT-4o / Hypothesizer →OpenAI	28% 1 report	DiscoveryWorld · Average (Easy) · Task completion[1]DiscoveryWorld · GPT-4o agent scaffolds	#2 of 3 in this displayed comparison; 10 pp below the leader. Source notes Source cell: 0.28
GPT-4o / Plan+Execute →OpenAI	18% 1 report	DiscoveryWorld · Average (Easy) · Task completion[1]DiscoveryWorld · GPT-4o agent scaffolds	#3 of 3 in this displayed comparison; 20 pp below the leader. Source notes Source cell: 0.18
GPT-4o / Hypothesizer →OpenAI	19% 1 report	DiscoveryWorld · Average (Normal) · Explanatory knowledge[1]DiscoveryWorld · GPT-4o agent scaffolds	Leads this displayed comparison. Source notes Source cell: 0.19
GPT-4o / ReACT →OpenAI	14% 1 report	DiscoveryWorld · Average (Normal) · Explanatory knowledge[1]DiscoveryWorld · GPT-4o agent scaffolds	#2 of 3 in this displayed comparison; 5 pp below the leader. Source notes Source cell: 0.14
GPT-4o / Plan+Execute →OpenAI	2% 1 report	DiscoveryWorld · Average (Normal) · Explanatory knowledge[1]DiscoveryWorld · GPT-4o agent scaffolds	#3 of 3 in this displayed comparison; 17 pp below the leader. Source notes Source cell: 0.02
GPT-4o / Plan+Execute →OpenAI	64% 1 report	DiscoveryWorld · Average (Normal) · Procedure[1]DiscoveryWorld · GPT-4o agent scaffolds	Leads this displayed comparison. Source notes Source cell: 0.64
GPT-4o / ReACT →OpenAI	63% 1 report	DiscoveryWorld · Average (Normal) · Procedure[1]DiscoveryWorld · GPT-4o agent scaffolds	#2 of 3 in this displayed comparison; 1 pp below the leader. Source notes Source cell: 0.63
GPT-4o / Hypothesizer →OpenAI	58% 1 report	DiscoveryWorld · Average (Normal) · Procedure[1]DiscoveryWorld · GPT-4o agent scaffolds	#3 of 3 in this displayed comparison; 6 pp below the leader. Source notes Source cell: 0.58
GPT-4o / Hypothesizer →OpenAI	23% 1 report	DiscoveryWorld · Average (Normal) · Task completion[1]DiscoveryWorld · GPT-4o agent scaffolds	Leads this displayed comparison. Source notes Source cell: 0.23
GPT-4o / Plan+Execute →OpenAI	18% 1 report	DiscoveryWorld · Average (Normal) · Task completion[1]DiscoveryWorld · GPT-4o agent scaffolds	#2 of 3 in this displayed comparison; 5 pp below the leader. Source notes Source cell: 0.18
GPT-4o / ReACT →OpenAI	18% 1 report	DiscoveryWorld · Average (Normal) · Task completion[1]DiscoveryWorld · GPT-4o agent scaffolds	#2 of 3 in this displayed comparison; 5 pp below the leader. Source notes Source cell: 0.18

Sources

DISCOVERYWORLD: A Virtual Environment for Developing and Evaluating Automated Scientific Discovery Agents — published result table ↗
Jansen, Peter, Côté, Marc-Alexandre, Khot, Tushar, Bransom, Erin, Mishra, Bhavana Dalvi, Majumder, Bodhisattwa Prasad et al. · 2024
All 8 authors
Jansen, Peter, Côté, Marc-Alexandre, Khot, Tushar, Bransom, Erin, Mishra, Bhavana Dalvi, Majumder, Bodhisattwa Prasad, Tafjord, Oyvind, Clark, Peter
Dates & evaluation details
DiscoveryWorld · Average (Easy) · Procedure · GPT-4o agent scaffolds · 2026-10-27
Published model comparison, 2024. DiscoveryWorld · GPT-4o agent scaffolds. Published difficulty-level mean over discovery tasks and parametric variants. GPT-4o is used by all three agent scaffolds. Procedural credit, task completion and explanatory knowledge are distinct metrics. Table 4; source precision and any reported uncertainty retained. These are source-reported observations, not evaluations run by RLTasks.
DiscoveryWorld · Average (Easy) · Task completion · GPT-4o agent scaffolds · 2026-10-27
Published model comparison, 2024. DiscoveryWorld · GPT-4o agent scaffolds. Published difficulty-level mean over discovery tasks and parametric variants. GPT-4o is used by all three agent scaffolds. Procedural credit, task completion and explanatory knowledge are distinct metrics. Table 4; source precision and any reported uncertainty retained. These are source-reported observations, not evaluations run by RLTasks.
DiscoveryWorld · Average (Easy) · Explanatory knowledge · GPT-4o agent scaffolds · 2026-10-27
Published model comparison, 2024. DiscoveryWorld · GPT-4o agent scaffolds. Published difficulty-level mean over discovery tasks and parametric variants. GPT-4o is used by all three agent scaffolds. Procedural credit, task completion and explanatory knowledge are distinct metrics. Table 4; source precision and any reported uncertainty retained. These are source-reported observations, not evaluations run by RLTasks.
DiscoveryWorld · Average (Normal) · Procedure · GPT-4o agent scaffolds · 2026-10-27
Published model comparison, 2024. DiscoveryWorld · GPT-4o agent scaffolds. Published difficulty-level mean over discovery tasks and parametric variants. GPT-4o is used by all three agent scaffolds. Procedural credit, task completion and explanatory knowledge are distinct metrics. Table 4; source precision and any reported uncertainty retained. These are source-reported observations, not evaluations run by RLTasks.
DiscoveryWorld · Average (Normal) · Task completion · GPT-4o agent scaffolds · 2026-10-27
Published model comparison, 2024. DiscoveryWorld · GPT-4o agent scaffolds. Published difficulty-level mean over discovery tasks and parametric variants. GPT-4o is used by all three agent scaffolds. Procedural credit, task completion and explanatory knowledge are distinct metrics. Table 4; source precision and any reported uncertainty retained. These are source-reported observations, not evaluations run by RLTasks.
DiscoveryWorld · Average (Normal) · Explanatory knowledge · GPT-4o agent scaffolds · 2026-10-27
Published model comparison, 2024. DiscoveryWorld · GPT-4o agent scaffolds. Published difficulty-level mean over discovery tasks and parametric variants. GPT-4o is used by all three agent scaffolds. Procedural credit, task completion and explanatory knowledge are distinct metrics. Table 4; source precision and any reported uncertainty retained. These are source-reported observations, not evaluations run by RLTasks.
DiscoveryWorld · Average (Challenge) · Procedure · GPT-4o agent scaffolds · 2026-10-27
Published model comparison, 2024. DiscoveryWorld · GPT-4o agent scaffolds. Published difficulty-level mean over discovery tasks and parametric variants. GPT-4o is used by all three agent scaffolds. Procedural credit, task completion and explanatory knowledge are distinct metrics. Table 4; source precision and any reported uncertainty retained. These are source-reported observations, not evaluations run by RLTasks.
DiscoveryWorld · Average (Challenge) · Task completion · GPT-4o agent scaffolds · 2026-10-27
Published model comparison, 2024. DiscoveryWorld · GPT-4o agent scaffolds. Published difficulty-level mean over discovery tasks and parametric variants. GPT-4o is used by all three agent scaffolds. Procedural credit, task completion and explanatory knowledge are distinct metrics. Table 4; source precision and any reported uncertainty retained. These are source-reported observations, not evaluations run by RLTasks.
DiscoveryWorld · Average (Challenge) · Explanatory knowledge · GPT-4o agent scaffolds · 2026-10-27
Published model comparison, 2024. DiscoveryWorld · GPT-4o agent scaffolds. Published difficulty-level mean over discovery tasks and parametric variants. GPT-4o is used by all three agent scaffolds. Procedural credit, task completion and explanatory knowledge are distinct metrics. Table 4; source precision and any reported uncertainty retained. These are source-reported observations, not evaluations run by RLTasks.

TRACE THE SCORE · Per-model notes use published breakdowns where available.

What it measures

Virtual interactive environment where agents formulate hypotheses, design controlled experiments, gather data, and discover scientific laws.

Reading the measurement

Ask whether the evaluation rewards recalling a fact, constructing a model, running an experiment or supporting a discovery. Those tasks require different evidence. For research claims, inspect reproducibility, uncertainty and what was actually checked.

Published evaluation

Published difficulty-level mean over discovery tasks and parametric variants. GPT-4o is used by all three agent scaffolds. Procedural credit, task completion and explanatory knowledge are distinct metrics.

Where this fits in an evaluation suite

A correct multiple-choice answer and a supported scientific contribution demand different evidence. Check whether the score rewards a proposal, an executed experiment, or a completed research target.

Start with a capability hypothesis, then choose a second test with a different failure mode. Keep the checkpoint, reasoning effort, agent tools, sampling policy, and scoring procedure fixed when measuring a training intervention. Record failures and missing outputs in the denominator.

A reproducible comparison

Publish the exact run configuration, item coverage, score units, and available uncertainty. If a source supplies only an aggregate or a single run, preserve that limit. A new model name or a rounded score alone does not establish a capability improvement.

PRIMARY EVIDENCE & FURTHER READING

DISCOVERYWORLD: A Virtual Environment for Developing and Evaluating Automated Scientific Discovery Agents — published result table ↗
Jansen, Peter, Côté, Marc-Alexandre, Khot, Tushar, Bransom, Erin, Mishra, Bhavana Dalvi, Majumder, Bodhisattwa Prasad et al. · 2024
All 8 authors
Jansen, Peter, Côté, Marc-Alexandre, Khot, Tushar, Bransom, Erin, Mishra, Bhavana Dalvi, Majumder, Bodhisattwa Prasad, Tafjord, Oyvind, Clark, Peter

Reviewed 2026-10-27. Explanation and interpretation by RLTasks. Benchmark ownership remains with its original authors.

Continue exploring

All scientific discovery benchmarks →

Sources

What it measures

Reading the measurement

Published evaluation

Where this fits in an evaluation suite

A reproducible comparison

Continue exploring

Terminal-Bench-Science 0.1

SciCode

GPQA Diamond

CritPt v2