Qwen2.5-Math-1.5B

1.54B parameters, text in, text out, 4K tokens of context. 1 provider serves it.

Verified 2026-10-25 Rank #251 in LLMs · score 85.8 method v1 Sources 3 Downloads 30d 231.2k · #238 of 619 tracked

Changed this week: 1 new quantized build on the Hub (17 in all).2026-10-24

Can I run it?0.9GB at INT4Estimatefits a 16 GB cardmethod

Can I use it?Apache 2.0Permissive; patent grantlicence file2026-10-24

Do people use it?231.2kdownloads, 30 days#238 of 619 we track · top 38%+2 likes, downloads −1% this week123 likesHugging Face2026-10-24

Worth reading

Picked by rule from 31 tracked pieces: the lab's own post, the most shared, a field report, the latest.

Latest · 2 years ago

Qwen2.5: A Party of Foundation Models

Qwen2.5 introduces dense decoder-only models pretrained on up to 18 trillion tokens, available in sizes from 0.5B to 72B parameters.

Qwen2.5-Coder-7B-Instruct outperforms many larger language models across a range of programming languages and tasks.
Qwen2.5-Math-72B-Instruct general performance surpasses both Qwen2-Math-72B-Instruct and GPT4-o.

qwenlm.github.io 2024-09-18 168 HN pts · 38 comments 1.7k words about the Qwen2.5 family family

25 best of 31 tracked

The 1 pick above, plus these 24. Top is the radar's score: discussion, source authority and the lab's own posts, fading with age.

Codekev: Jev-like model built on Qwen2.5-0.5Bgithub.com · 2026-09-19
ArticleQwen2.5-VL-32B: Smarter and Lighterqwenlm.github.io · 2025-03-24 · official
ArticleQwen2.5-1M: Deploy your own Qwen with context length up to 1M tokensqwenlm.github.io · 2025-01-26 · official
ArticleQwen2.5-Max: Exploring the intelligence of large-scale MoE modelqwenlm.github.io · 2025-01-28 · official
ArticleQwen2.5-Coder-32B is an LLM that can code well that runs on my Macsimonwillison.net · 2024-11-13
ArticleShow HN: Fine-tuned Qwen2.5-7B on 100 films for probabilistic story graphscinegraphs.ai · 2026-02-08
ArticleQwen2.5-7B-Instruct-1M and Qwen2.5-14B-Instruct-1Msimonwillison.net · 2025-01-26
PaperQwen2.5-Math Technical Report: Toward Mathematical Expert Model via Self-Improvementarxiv.org · undated
ArticleQwen2.5-Coder Beats GPT-4o and Claude 3.5 Sonnet in Codingqwenlm.github.io · 2024-11-11 · official
ArticleCivic-SLM is a domain-specialized fine-tune of Qwen2.5-7B for U.S. govt dataitsmeduncan.com · 2026-04-25
ArticleQwen2.5-VLqwenlm.github.io · 2025-01-27 · official
CodeQwen2.5-Coder-32B with coding abilities matching those of GPT-4ogithub.com · 2024-11-09
ArticleQwen2.5 Omni: See, Hear, Talk, Write, Do It Allqwenlm.github.io · 2025-03-26 · official
ArticleAlibaba Qwen2.5-Omni-7B: Open Source End-to-End Multimodal AI Modelalizila.com · 2025-03-27
Code用评测代码测试Qwen2.5-Math-1.5B 结果和report 的 ...github.com · undated
ArticleQwen2.5-VL-32B: Smarter and Lightersimonwillison.net · 2025-03-25
ArticleQwen2.5-Math: The world's leading open-sourced mathematical LLMs | Qwenqwenlm.github.io · 2024-09-18 · official
ArticleQwen2.5-Max: The Next Leap in AI Innovationmedium.com · 2025-01-29
ArticleQwen2.5-Omni Technical Reporthuggingface.co · 2025-03-30
PaperQwen2.5-Coder Technical Reportarxiv.org · 2024-11-11
DiscussionQwen2.5-coder on the events occurring during the Tiananmen Square protestsnews.ycombinator.com · 2024-11-20
PaperQwen2.5 Technical Reportarxiv.org · 2025-01-03
CodeQwen2.5-VL: State-of-the-art multimodal LLMgithub.com · 2025-01-30
DiscussionNew Qwen2.5-Max Outperforms DeepSeek V3 in Benchmarkstwitter.com · 2025-01-28

Show all 24

Where to run it

Live hosted endpoints, cheapest first, USD per million tokens. Ties break on output price, then provider name.

#	Cheapest live row	In /1M	Out /1M	Context	Quant	Seen
1	Featherless	—	—	—	—	Oct 25

All 1 provider →

Where it's used

Hugging Face does not say who downloads a model or with what. These are the proxies: API providers listing it, quantized builds, fine-tunes/adapters/merges, and Spaces that load it.

Served by 1 API provider; 17 quantized builds for local use; a big base for builders (381 fine-tunes, adapters and merges).

As a service1Featherless
On own hardware17quantized builds
Built on381330 fine-tunes · 20 adapters · 31 merges
In apps12Hugging Face Spaces

Top Spaces running it: hfmlsoc/different-flops ♥ 11 · trl-lib/recommend-vllm-memory ♥ 6 · adi2606/Math_Wizard ♥ 2 · hikefps/openenv-nifty500-stock-agent-rl ♥ 1

Will it fit?

Weights only, an estimate: leave room for the KV cache. Method

Fits a 16 GB card at INT4

About 0.9 GB at INT4, 1.9 GB at FP8, 3.7 GB at BF16: weights only, so leave room for the KV cache.

Estimate params × bytes per param × 1.2 · how we estimate

The story so far

Weights, every dated article, and repo commits on one rail.

Most coverage came in the first two weeks

Teal is the weights (2024-09-16), amber triangles are 5 dated articles (each links to it), grey ticks are repo commits.

What people built on it

398 derivatives on the Hub. Which build to run, by runtime: the most downloaded in each class.

398 derivatives on the Hub, most of them fine-tunes

From the Hugging Face model tree; quantized builds in teal, fine-tunes, merges and adapters in greys.

GGUF mradermacher/Qwen2.5-Math-1.5B-i1-GGUF for llama.cpp, Ollama, LM Studio 13.6k downloads MLX mlx-community/Qwen2.5-Math-1.5B-Instruct-8bit for Apple silicon 471 downloads

All variants →

Reference

Everything else, folded: the full tables, every source.

What the card says

1 cited sentence

Summary generated, cited

The model is designed to solve English and Chinese math problems using Chain-of-Thought and Tool-integrated Reasoning.^[1]

“support using both CoT and Tool-integrated Reasoning (TIR…”

Generated 2026-10-07 15:07 UTC by parasail-gemma-4-26b-a4b-it; each numeral points at the fact it rests on. Not a verdict.

Use-case evidence

4 use cases

What it is good at 4 use cases

Labels are evidence, not verdicts: measured a third-party leaderboard row for this model; author-reported a number the model author published in the card or eval files; structural a deterministic signal from the name, config or chat template, or a decision-model estimate. Each chip links to its source. All use cases.

Facts

22 fields

Listed in LLMs

Facts

Qwen2.5-Math-1.5B has 1.54B parameters. (safetensors metadata, observed 2026-10-24)
Context length is 4,096 tokens (4K). (config.json max_position_embeddings, observed 2026-10-24)
License: Apache 2.0 (friction: none). Permissive; patent grant. (license file, observed 2026-10-24)
Created 2024-09-16, last modified 2024-09-23. (HF API, observed 2026-10-24)

Pipeline tag	text-generation
Library	transformers
Architectures	Qwen2ForCausalLM
Model type	qwen2
Dtype	bfloat16
Params by dtype	BF16: 1,543,714,304
Size band	S
MoE	no
Context length	4,096 (config.json max_position_embeddings)
Languages	en
Papers	arXiv:2409.12122
License	Apache 2.0 (apache-2.0)
Hub snapshot	2026-10-25

Servability

vLLM · SGLang

Servability

vLLM: supported · Qwen2ForCausalLM · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · Qwen2ForCausalLM · 773cb938fe9484a8f7e7e46fb1b1fe96412c59ed

vLLM registry (7050ef881ff0835f6890a47db1859a24c65481d6) · SGLang registry (773cb938fe9484a8f7e7e46fb1b1fe96412c59ed) · Matched by architecture class name against each project's registry on main; 'supported' means the architecture is registered, not that this checkpoint was tested.

Hardware fit

3 estimates

Hardware fit ESTIMATE

Weights-only memory for Qwen2.5-Math-1.5B: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.

BF16 (2 bytes/param)	3.7 GB · Edge (<4 GB)
FP8 (1 byte/param)	1.9 GB · Edge (<4 GB)
INT4 (0.5 bytes/param)	0.9 GB · Edge (<4 GB)

Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra.

All variants

40 variants

Lineage

Fine-tuned from Qwen/Qwen2.5-1.5B (same organization) Family: Qwen/Qwen2.5 398 derivatives on the Hub

Base model: Qwen/Qwen2.5-1.5B

finetune: 330quantized: 17adapter: 20merge: 31

Top variants (40 shown, ranked by downloads)

Variant	Relation	Quant	Params	Band	License	Author	Likes	Downloads
TIGER-Lab/general-verifier	finetune	—	1.54B	S	apache-2.0	TIGER-Lab	21	103.6k
WeiboAI/VibeThinker-1.5B	finetune	—	1.78B	S	mit	WeiboAI	559	72.1k
unsloth/Qwen2.5-Math-1.5B	finetune	—	1.54B	S	apache-2.0	unsloth	4	56.5k
nvidia/OpenMath-Nemotron-1.5B	finetune	—	1.54B	S	cc-by-4.0	nvidia	36	53.9k
unsloth/Qwen2.5-Math-1.5B-bnb-4bit	quantized	-bnb	1.58B	S	apache-2.0	unsloth	3	49k
sail/Qwen2.5-Math-1.5B-Oat-Zero	finetune	—	1.54B	S	apache-2.0	sail	4	31k
mradermacher/Qwen2.5-Math-1.5B-i1-GGUF	quantized	GGUF	—	—	apache-2.0	mradermacher	0	13.6k
WiroAI/WiroAI-Finance-Qwen-1.5B	finetune	—	1.78B	S	apache-2.0	WiroAI	25	9.5k
Mungert/OpenMath-Nemotron-1.5B-GGUF	quantized	GGUF	—	—	cc-by-4.0	Mungert	0	9.5k
QuantFactory/Qwen2.5-Math-1.5B-Instruct-GGUF	quantized	GGUF	—	—	apache-2.0	QuantFactory	1	8.9k
Mungert/VibeThinker-1.5B-GGUF	quantized	GGUF	—	—	mit	Mungert	4	8.2k
mradermacher/Qwen2.5-Math-1.5B-GGUF	quantized	GGUF	—	—	apache-2.0	mradermacher	0	7.3k
Elliott/LUFFY-Qwen-Math-1.5B-Zero	finetune	—	1.54B	S	mit	Elliott	0	5.9k
ypwang61/One-Shot-RLVR-Qwen2.5-Math-1.5B-pi1	finetune	—	1.78B	S	apache-2.0	ypwang61	0	4.4k
ypwang61/One-Shot-RLVR-Qwen2.5-Math-1.5B-pi13	finetune	—	1.78B	S	apache-2.0	ypwang61	0	3.2k
featherless-ai-quants/Qwen-Qwen2.5-Math-1.5B-GGUF	quantized	GGUF	—	—	—	featherless-ai-quants	0	2.1k
ypwang61/One-Shot-RLVR-Qwen2.5-Math-1.5B-7.5k-MATH	finetune	—	1.78B	S	apache-2.0	ypwang61	0	2k
DevQuasar-7/Qwen.Qwen2.5-Math-1.5B-GGUF	quantized	GGUF	—	—	—	DevQuasar-7	0	1.9k
RecursiveMAS/Sequential-Light-Solver-Qwen2.5-Math-1.5B	finetune	—	1.54B	S	mit	RecursiveMAS	0	1.6k
ypwang61/One-Shot-RLVR-R1-Distill-1.5B-pi1	finetune	—	1.78B	S	apache-2.0	ypwang61	0	719
ypwang61/One-Shot-RLVR-R1-Distill-1.5B-1.2k-dsr-sub	finetune	—	1.78B	S	apache-2.0	ypwang61	0	618
zbeeb/Qwen2.5-Math-1.5B-Base-GRPO	finetune	—	1.78B	S	apache-2.0	zbeeb	1	559
ypwang61/One-Shot-RLVR-R1-Distill-1.5B-16-shot	finetune	—	1.78B	S	apache-2.0	ypwang61	0	599
lucaswychan/Qwen2.5-Math-1.5B-Reasoning-Embedding	finetune	—	1.54B	S	apache-2.0	lucaswychan	0	592
zbeeb/Qwen2.5-Math-1.5B-GRPO-Staleness-2	finetune	—	1.54B	S	apache-2.0	zbeeb	0	587
daviddavidlu/DAPO-with-prompt-augmentation-step2820	finetune	—	1.78B	S	apache-2.0	daviddavidlu	0	572
ypwang61/One-Shot-RLVR-R1-Distill-1.5B-4-shot	finetune	—	1.78B	S	apache-2.0	ypwang61	0	556
daviddavidlu/DAPO-with-prompt-augmentation-step2480	finetune	—	1.78B	S	apache-2.0	daviddavidlu	0	531
daviddavidlu/DAPO-with-prompt-augmentation-step2720	finetune	—	1.78B	S	apache-2.0	daviddavidlu	0	526
zbeeb/Qwen2.5-Math-1.5B-GRPO-Staleness-6	finetune	—	1.54B	S	apache-2.0	zbeeb	0	522
Habibur2/Qwen2.5-Math-1.5B-GGUF	quantized	GGUF	—	—	apache-2.0	Habibur2	2	421
SandLogicTechnologies/Qwen2.5-Math-1.5B-GGUF	quantized	GGUF	—	—	apache-2.0	SandLogicTechnologies	0	508
mlx-community/Qwen2.5-Math-1.5B-Instruct-8bit	finetune	-8bit	1.54B	S	apache-2.0	mlx-community	0	471
ypwang61/One-Shot-RLVR-Qwen2.5-Math-1.5B-1.2k-dsr-sub	finetune	—	1.78B	S	apache-2.0	ypwang61	0	467
mlx-community/Qwen2.5-Math-1.5B-Instruct-bf16	finetune	-bf16	1.54B	S	apache-2.0	mlx-community	0	464
zhaohq/PureRL-1.5B-v6g-A-lam01-sigmoid-maskoff	finetune	—	1.78B	S	—	zhaohq	0	460
chinmaydk99/Qwen2.5-1.5B-Math-Open-R1-Distill	finetune	—	1.54B	S	—	chinmaydk99	0	458
zhaohq/PureRL-1.5B-v6g-B-lam03-sigmoid-maskoff	finetune	—	1.78B	S	—	zhaohq	0	452
Jongbin-kr/qwen25-math15b-cvar-grpo-acemath-full3663-cl768-l005	adapter	—	—	—	—	Jongbin-kr	0	425
JaeWooShin/qwen2.5-math-1.5b-fewgt-u1-n256-inverted-lam1-s0	finetune	—	1.54B	S	apache-2.0	JaeWooShin	0	424

Repository changelog

10 commits

changed in the last 7 daysOther: 2026-10-24all changes

Repository changelog 10 commits

Date	Commit	SHA
2024-09-23	Update README.md	4a83ca6
2024-09-20	Upload ./tokenizer_config.json with huggingface_hub	6d29d57
2024-09-19	Upload ./tokenizer_config.json with huggingface_hub	32c01b0
2024-09-18	Update README.md	e0a5540
2024-09-18	Create LICENSE	0b0c5af
2024-09-18	Create README.md	13adcdb
2024-09-17	Update tokenizer_config.json	37193e8
2024-09-16	Delete model.safetensors.index.json	a3af4eb
2024-09-16	Upload folder using huggingface_hub	b53e4e3
2024-09-16	initial commit	d9401d4

All providers

1 provider · 1 endpoint

Where to run 1 providers · 1 endpoints

Hosted endpoints observed for Qwen2.5-Math-1.5B, sorted by input price (USD per million tokens). Each row names its observer and date; prices change daily, so check the source before quoting.

Provider	Model ref	Context	Quant	$/M in	$/M out	tok/s	TTFT	Tools	Status	Observed
Featherless hf	Qwen/Qwen2.5-Math-1.5B	—	—	—	—	—	—	—	live	2026-10-25

Articles and coverage

31 pieces

Ranked by source signals (HN, Lobsters, authority domains, official posts) and recency. Summaries are generated from a stored copy of the article text; every sentence cites a span of that text, and nothing is shown that failed the citation check.

Official2

Qwen2.5: A Party of Foundation Models

qwenlm.github.io2024-09-18officialauthorityHN front pageabout the Qwen2.5 familyHN 168 pts · 38 comments1,744 words

Qwen2.5 introduces dense decoder-only models pretrained on up to 18 trillion tokens, available in sizes from 0.5B to 72B parameters. Most variants are licensed under Apache 2.0, excluding the 3B and 72B models, and support a 128K context window with 8K generation limits. Specialized versions for coding and mathematics offer enhanced performance on benchmarks like HumanEval and MATH.written, reviewed, cited

Qwen2.5-Coder-7B-Instruct outperforms many larger language models across a range of programming languages and tasks.
Qwen2.5-Math-72B-Instruct general performance surpasses both Qwen2-Math-72B-Instruct and GPT4-o.

Qwen2.5-Math: The world's leading open-sourced mathematical LLMs | Qwen

qwenlm.github.io2024-09-18officialauthorityabout the Qwen2.5 family

Analysis6

Qwen2.5-Math Technical Report: Toward Mathematical Expert Model via Self-Improvement

arxiv.orgauthorityabout the Qwen2.5 family12,642 words

用评测代码测试Qwen2.5-Math-1.5B 结果和report 的 ...

github.comauthority

Qwen2.5 Technical Report

arxiv.org2025-01-03authorityabout the Qwen2.5 family

Qwen2.5-Math-1.5B Model

emergentmind.com

QwenLM/Qwen2.5-Math

github.com2024-08-08authorityabout the Qwen2.5 family

Show HN: I made a free study tool powered by Qwen2.5

freeflashcardgenerator.com2025-01-28about the Qwen2.5 familyHN 3 pts · 0 comments

All coverage tracked for this model (31) — append-only, nothing is ever removed

Title	Source	Published	First seen	Signals
Qwen2.5-Math Technical Report: Toward Mathematical Expert Model via Self-Improvement	arxiv.org	—	2026-10-07	exa
用评测代码测试Qwen2.5-Math-1.5B 结果和report 的 ...	github.com	—	2026-10-07	exa
Qwen2.5-Math-1.5B Model	emergentmind.com	—	2026-10-07	exa
kev: Jev-like model built on Qwen2.5-0.5B	github.com	2026-09-19	2026-10-07	hn
Civic-SLM is a domain-specialized fine-tune of Qwen2.5-7B for U.S. govt data	itsmeduncan.com	2026-04-25	2026-10-07	hn
Show HN: Fine-tuned Qwen2.5-7B on 100 films for probabilistic story graphs	cinegraphs.ai	2026-02-08	2026-10-07	hn
Qwen2.5-Omni Technical Report	huggingface.co	2025-03-30	2026-10-07	hn
Alibaba Qwen2.5-Omni-7B: Open Source End-to-End Multimodal AI Model	alizila.com	2025-03-27	2026-10-07	hn
Qwen2.5 Omni: See, Hear, Talk, Write, Do It All	qwenlm.github.io official	2025-03-26	2026-10-07	hn
Qwen2.5-VL-32B: Smarter and Lighter	simonwillison.net	2025-03-25	2026-10-07	hn
Qwen2.5-VL-32B: Smarter and Lighter	qwenlm.github.io official	2025-03-24	2026-10-07	hn
Qwen2.5-Plus	chat.qwenlm.ai	2025-01-31	2026-10-07	hn
Qwen2.5-VL: State-of-the-art multimodal LLM	github.com	2025-01-30	2026-10-07	hn
Qwen2.5-Max: The Next Leap in AI Innovation	medium.com	2025-01-29	2026-10-07	hn
New Qwen2.5-Max Outperforms DeepSeek V3 in Benchmarks	twitter.com	2025-01-28	2026-10-07	hn
Qwen2.5-Max: Exploring the intelligence of large-scale MoE model	qwenlm.github.io official	2025-01-28	2026-10-07	hn
Show HN: I made a free study tool powered by Qwen2.5	freeflashcardgenerator.com	2025-01-28	2026-10-07	hn
Qwen2.5-VL	qwenlm.github.io official	2025-01-27	2026-10-07	hn
Qwen2.5-7B-Instruct-1M and Qwen2.5-14B-Instruct-1M	simonwillison.net	2025-01-26	2026-10-07	hn
Qwen2.5-1M: Deploy your own Qwen with context length up to 1M tokens	qwenlm.github.io official	2025-01-26	2026-10-07	hn
Qwen2.5 Technical Report	arxiv.org	2025-01-03	2026-10-07	exa
Qwen2.5-coder on the events occurring during the Tiananmen Square protests	news.ycombinator.com	2024-11-20	2026-10-07	hn
Qwen2.5-Coder-32B is an LLM that can code well that runs on my Mac	simonwillison.net	2024-11-13	2026-10-07	hn
Qwen2.5-Coder Technical Report	arxiv.org	2024-11-11	2026-10-07	hn
Qwen2.5 Coder	ollama.com	2024-11-11	2026-10-07	hn
Qwen2.5-Coder Beats GPT-4o and Claude 3.5 Sonnet in Coding	qwenlm.github.io official	2024-11-11	2026-10-07	hn
Qwen2.5-Coder-32B with coding abilities matching those of GPT-4o	github.com	2024-11-09	2026-10-07	hn
Guide how to use a local LLM Qwen2.5-coder-7B as an autocompelte in VSCode	eamag.me	2024-10-29	2026-10-07	hn
Qwen2.5: A Party of Foundation Models	qwenlm.github.io official	2024-09-18	2026-10-07	hn
Qwen2.5-Math: The world's leading open-sourced mathematical LLMs \| Qwen	qwenlm.github.io official	2024-09-18	2026-10-07	exa
QwenLM/Qwen2.5-Math	github.com	2024-08-08	2026-10-07	exa

Papers

1 paper

Papers 1

arXiv links found in the model card and its tags. "Peer-reviewed" is shown only with evidence: a published venue in OpenAlex (measured) or the authors' own journal reference on arXiv (reported).

Qwen2.5-Math Technical Report: Toward Mathematical Expert Model via Self-Improvement this model

An Yang, Beichen Zhang, Binyuan Hui +13 · 2024-09-18 · PDF · HF paper page (5 upvotes)

Similar models

3 models

Similar models

Qwen2-1.5B-Instruct · 1.54B
same size band (S), same task (text-generation), permissive license
Qwen2-1.5B · 1.54B
same size band (S), same task (text-generation), permissive license
ReaderLM-v2 · 1.54B
same size band (S), same task (text-generation)

Sources

3 sources

Sources

Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:37 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:45 UTC · Summary: 2026-10-07 15:07 UTC · Last verified 2026-10-25

Follow Qwen2.5-Math-1.5B

Watch this model

/feeds/models/Qwen--Qwen2.5-Math-1.5B.xml(RSS 2.0) · every change to Qwen2.5-Math-1.5B
/feeds/families/Qwen--Qwen2.5.xml(RSS 2.0) · every change across the Qwen2.5 family

Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.