DeepSeekMath-V2
Shares rank 2 of 3 math reasoning models by recommenders
Free open model praised for strong math proofs and problem solving; main caveat is it is weak outside math.
Free open model praised for strong math proofs and problem solving; main caveat is it is weak outside math.
16 people gave an opinion
- recommend it
- 10
- criticize it
- 6
- approval share
- 63%
Recommended in 6 of 13 months
- November 2025: recommended
- December 2025: recommended
- January 2026: recommended
- February 2026: recommended
- March 2026: not recommended
- April 2026: recommended
- May 2026: not recommended
- June 2026: not recommended
- July 2026: not recommended
- August 2026: not recommended
- September 2026: not recommended
- October 2026: recommended
- November 2026: not recommended (month in progress)
Why it doesn’t carry the badge
- 10 people recommend itNeeds 10 or more, met.
- 1.6 recommenders for every criticNeeds 2 to 1 or better, not met.
- Recommended in 6 of 13 monthsNeeds 4 or more, met.
The same three rules apply to every product, and the badge can’t be bought. How the badge works
What people say about it
The reasons people give, with the number of comments that give each one. Open a reason to read what they said.
Why people recommend it
Each mark is one comment
Outperforms larger models on math
“DeepSeek R1 and DeepSeekMath were groundbreaking models which were substantially stronger than any models available for distillation at the time they were developed”
Comment from 24 Oct 2026 in “Kimi K3.1 Pro and K3.1 Pro Highspeed” “They have had the best math models for about a year most folks just didn't know about it.”
Comment from 24 Apr 2026 in “DeepSeek v4” “It was a tiny model that outperformed huge then-SOTA LLMs like Claude 3 Opus in math”
Comment from 8 Dec 2025 in “Google Titans architecture, helping AI have long-term memory” Strong competition math results
“Is everyone just glossing over the first place score of 118/120 on the Putnam?”
Comment from 27 Nov 2025 in “DeepSeekMath-V2: Towards Self-Verifiable Mathematical Reasoning [pdf]” “the model achieved nearly 99% accuracy”
Comment from 28 Nov 2025 in “DeepSeekMath-V2: Towards Self-Verifiable Mathematical Reasoning [pdf]”
What people warn about
Each mark is one comment
Self-graded results, possible contamination
“they are using their own verifier as apanem of experts, with their own model trained specifically to satisfy this verifier”
Comment from 28 Nov 2025 in “DeepSeekMath-V2: Towards Self-Verifiable Mathematical Reasoning [pdf]” “They do not mention doing any benchmark decontamination or excluding 2024/2025 competition problems from training.”
Comment from 1 Dec 2025 in “DeepSeekMath-V2: Towards Self-Verifiable Mathematical Reasoning” “it's not a real score; they judged the results themselves and Putnam judges are notoriously tough”
Comment from 28 Nov 2025 in “DeepSeekMath-V2: Towards Self-Verifiable Mathematical Reasoning [pdf]” Narrow, overeager math focus
“It's so eager to solve math problems, that it even comes up with random ones if you prompt it with "Hello".”
Comment from 1 Dec 2025 in “DeepSeekMath-V2: Towards Self-Verifiable Mathematical Reasoning” “This model can’t be used for say questions on biology or history.”
Comment from 1 Dec 2025 in “DeepSeekMath-V2: Towards Self-Verifiable Mathematical Reasoning”
Other math reasoning models people recommend
The two most recommended other math reasoning models, with the same counts for each, so you can compare like for like. The tick on each bar marks the 2 to 1 line the badge needs.
| Product | Recommend | Criticize | Approval | Months | Owned | Badge |
|---|---|---|---|---|---|---|
| Leanstral by Mistral | 15 | 12 | 56% | 2 | — | — |
| DeepSeekMath-V2This page | 10 | 6 | 63% | 6 | — | — |
| Harmonic Aristotle | 10 | 2 | 83% | 4 | — | Approved |
What people said
16 of the 17 comments that mention it, newest first. Each links to the original comment.
“DeepSeek R1 and DeepSeekMath were groundbreaking models which were substantially stronger than any models available for distillation at the time they were developed”
“your point is super valid and I agree with you”
“They have had the best math models for about a year most folks just didn't know about it.”
“Deepseek wrote a decent paper on this”
“if you want to do math proofs use AI built for proof”
“It was a tiny model that outperformed huge then-SOTA LLMs like Claude 3 Opus in math”
“They do not mention doing any benchmark decontamination or excluding 2024/2025 competition problems from training.”
“This model can’t be used for say questions on biology or history.”
“It's so eager to solve math problems, that it even comes up with random ones if you prompt it with "Hello".”
“it's literally a compiled binary, without even the source data”
“they are using their own verifier as apanem of experts, with their own model trained specifically to satisfy this verifier”
“it's not a real score; they judged the results themselves and Putnam judges are notoriously tough”
“having a free-of-charge maths genius available 24/7 has broad potential”
“the model achieved nearly 99% accuracy”
“Is everyone just glossing over the first place score of 118/120 on the Putnam?”
“Exciting stuff from a fantastic team.”
Counts come from 1,795,640 Hacker News comments, November 2025 to November 2026. How we count. Report a mistake.