ToRA makes tool-integrated reasoning a competitive open math-agent paradigm

The agents learn interactive tool-use trajectories involving computation libraries and symbolic solvers rather than relying on text-only chain-of-thought.

What happened?

ToRA-7B reached 44.6% on MATH and ToRA-Code-34B exceeded 50%, using trajectories that interleave reasoning with computational tools.

Why it matters

Tool-integrated reasoning became a reusable pattern for later math systems that need exact arithmetic or programmatic checking.

Significance rationale: ToRA demonstrated that language reasoning interleaved with executable tools could substantially improve difficult mathematical problem solving in open models.

Historical context

Primary evidence

Evidence & status at a glance

Significance
H2 · Field milestone
Verification
Paper released
Evidence level
E3 · Inspectable artifact
AI role
AI primary, human verified
Formal assurance
Artifact available

How should these labels be read? →

Contribution & verification

Mathematical novelty
Benchmark result
Interface
Informal, Symbolic, Numeric, Hybrid
AI contribution
The model decides when to reason in language and when to call external computational tools.
Human contribution
Researchers curated tool-use trajectories, trained the agents and designed evaluations.
Verification
Paper released · E3 · Inspectable artifact · Artifact available
Last reviewed
2026-09-11

Technical details

Systems
tora
Organizations
microsoft-research, tsinghua-university
Problems
math-benchmark
Methods
tool-integrated-reasoning, code-execution, symbolic-tools

Sources & artifacts

Verification history

  • 2023-09-29 · Paper released · E2 · Research record

    The Chronicle records this event from the cited primary or official source; the status does not imply independent Chronicle verification.

  • 2026-09-11 · Paper released · E3 · Inspectable artifact

    This corpus refresh added a directly inspectable public artifact maintained by the authors or releasing organization. E3 records artifact-level inspectability and does not imply independent verification of the underlying scientific or mathematical claim.

Corrections

No corrections recorded.