DeepSeek
xAI
MistralResearch retrieval or analysis?
Search appears as an explicit task for only one model on this shortlist. For every other model, retrieval tooling and citation behavior require separate verification.
Purpose guide · Reasoning and research
Which AI models belong on a reasoning or research shortlist?
Catalog current through July 14, 2026
5 models
Five source-backed models whose Epoch records explicitly name reasoning, proof, or search work. This is a shortlist, not a performance ranking.
Selection rule
Included when the Epoch record explicitly lists mathematical reasoning, quantitative reasoning, theorem proving, proof generation, proof verification, or search. A matching label establishes task relevance only, not quality.
Task-fit shortlist
Evidence first. Your evaluation next.
Ordered by release date for readability, newest first. The order is not a score or recommendation.
- 01
- Matching evidence
- Mathematical reasoning · Theorem proving · Proof generation · Proof verification
- Domain
- Language
- Access
- Open weights (unrestricted); open weights
- 02
- Matching evidence
- Mathematical reasoning · Quantitative reasoning
- Domain
- Language
- Access
- API access; closed weights
- 03
- Matching evidence
- Search · Quantitative reasoning
- Domain
- Language · Multimodal · Vision
- Access
- API access; closed weights
- 04
- Matching evidence
- Quantitative reasoning
- Domain
- Language
- Access
- Open weights (unrestricted); open weights
- 05
- Matching evidence
- Mathematical reasoning
- Domain
- Multimodal · Language · Vision
- Access
- Open weights (non-commercial); open weights
Decision criteria
Questions the source can help you ask.
These checks narrow an evaluation plan. They do not replace hands-on testing.
General reasoning or formal proof?
Use the task labels to distinguish quantitative reasoning from theorem and proof work. Do not assume one label covers the other.
What must you test yourself?
Evaluate factual accuracy, source fidelity, calibration, tool use, latency, context handling, and cost on your own workload. This catalog does not measure them.
Evidence boundary
What this guide does not know.
No task-level quality, price, context window, latency, or reliability measurement is present in the Epoch catalog. We therefore publish no ranking or “best” label.
Catalog source: Epoch AI Data on AI Models, retrieved 2026-07-14, CC-BY. Artifact sha256:3c663b17ec989f00ef5f994214b878a914118e2d1fb6077e23329a318209eb86.
Each listed model links to its model-level source above. See the publication source ledger and methodology for the wider evidence policy.