# deepseek-ai/DeepSeek-Math-V2

Repository: https://github.com/deepseek-ai/DeepSeek-Math-V2
Canonical: https://ross.abutalabs.com/products/deepseek-math-v2
Language: Python
License: Apache-2.0
License Family: permissive
Last push: 2025-12-01T07:01:42+00:00

## Health v2 (maintenance only)
Score: 41/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 55, release rhythm 35, longevity 19
- inputs: {"age_days": 279, "days_push": 275, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1603, forks 153 (observed 2026-08-28T04:05:09.898313+00:00)

## What it is
DeepSeekMath-V2 is a research release from DeepSeek AI presenting an open large language model trained for self-verifiable mathematical reasoning and theorem proving. The repository provides model weights, a technical report, and inference/evaluation code under an Apache-2.0 license.

## Use cases
- run a model that solves olympiad math problems with rigorous proofs
- download open weights for theorem-proving research
- evaluate LLM mathematical reasoning on competition benchmarks
- study verifier-based reward model training for math
- generate step-by-step derivations instead of just final answers
- fine-tune or build on an open math reasoning model

## When to choose
- you need an open-weights LLM specialized in rigorous mathematical reasoning and proof generation
- you are researching self-verification or verifier-as-reward-model training
- you want to benchmark theorem-proving capability

## When to avoid
- you need a general-purpose chat or coding assistant
- you lack GPU resources to run large model weights
- you need a hosted API rather than self-hosted weights

## Facets
- artifact type: learning-resource
- maturity: active
- function: machine-learning, deep-learning, llm-training, nlp
- domain: large-language-models, artificial-intelligence, mathematics
- platform: python
- tags: model-weights, mathematical-reasoning, theorem-proving, self-verification, reinforcement-learning, research, huggingface, natural-language-processing, gpu, linux

## Member repositories
- deepseek-ai/DeepSeek-Math-V2 (main) score 41

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:09.898313+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T03:52:22.923951+00:00, confidence not recorded.
  - readme: https://github.com/deepseek-ai/DeepSeek-Math-V2 (fetched 2026-08-28T04:05:09.898313+00:00, sha b476b45b278d)
- Data as of 2026-08-30T08:39:29.467469+00:00.