Model Follower.

Up to date

Rebuilt
2026-10-25 06:54 UTC
Release radar
117 on watch
Hub snapshot
2026-10-25
Re-checked
daily · every number sourced and dated

Models › Qwen2.5 family

Qwen2.5-14B-Instruct-1M

Hugging Face ↗ Follow changes

Qwen Apache 2.0 out 21 months ago

14.8B parameters, text in, text out, 986K tokens of context. 1 provider serves it.

Verified 2026-10-25 Rank #461 in LLMs · score 77.7 method v1 Sources 3 Downloads 30d 9.7k · #509 of 619 tracked

Can I run it?8.9GB at INT4Estimatefits a 16 GB cardmethod
Can I use it?Apache 2.0Permissive; patent grantlicence file2026-10-24
Do people use it?9.7kdownloads, 30 days#509 of 619 we track · below the median+2 likes, downloads −60% this week347 likesHugging Face2026-10-24

Worth reading

Picked by rule from 25 tracked pieces: the lab's own post, the most shared, a field report, the latest.

The lab's own post

Qwen2.5-1M: Deploy your own Qwen with context length up to 1M tokens

QwenLM releases open-source Qwen2.5-1M models supporting up to one million token contexts.

  • Training begins with a pre-trained checkpoint that originally had a 4K token context length before expansion.
  • The 7B model exhibits minor errors when retrieving information from documents containing up to 1M tokens.
qwenlm.github.io 2025-01-26 307 HN pts · 108 comments 1.6k words
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 25 tracked

The 3 picks above, plus these 22. Top is the radar's score: discussion, source authority and the lab's own posts, fading with age.

  1. Codekev: Jev-like model built on Qwen2.5-0.5Bgithub.com · 2026-09-19
  2. ArticleQwen2.5-VL-32B: Smarter and Lighterqwenlm.github.io · 2025-03-24 · official
  3. ArticleQwen2.5-Max: Exploring the intelligence of large-scale MoE modelqwenlm.github.io · 2025-01-28 · official
  4. ArticleQwen2.5-Coder-32B is an LLM that can code well that runs on my Macsimonwillison.net · 2024-11-13
  5. ArticleShow HN: Fine-tuned Qwen2.5-7B on 100 films for probabilistic story graphscinegraphs.ai · 2026-02-08
  6. ArticleQwen2.5-Coder Beats GPT-4o and Claude 3.5 Sonnet in Codingqwenlm.github.io · 2024-11-11 · official
  7. ArticleCivic-SLM is a domain-specialized fine-tune of Qwen2.5-7B for U.S. govt dataitsmeduncan.com · 2026-04-25
  8. ArticleQwen2.5-VLqwenlm.github.io · 2025-01-27 · official
  9. CodeQwen2.5-Coder-32B with coding abilities matching those of GPT-4ogithub.com · 2024-11-09
  10. ArticleQwen2.5 Omni: See, Hear, Talk, Write, Do It Allqwenlm.github.io · 2025-03-26 · official
  11. ArticleAlibaba Qwen2.5-Omni-7B: Open Source End-to-End Multimodal AI Modelalizila.com · 2025-03-27
  12. ArticleQwen2.5-VL-32B: Smarter and Lightersimonwillison.net · 2025-03-25
  13. ArticleQwen2.5-Max: The Next Leap in AI Innovationmedium.com · 2025-01-29
  14. ArticleQwen2.5-Omni Technical Reporthuggingface.co · 2025-03-30
  15. PaperQwen2.5-Coder Technical Reportarxiv.org · 2024-11-11
  16. DiscussionQwen2.5-coder on the events occurring during the Tiananmen Square protestsnews.ycombinator.com · 2024-11-20
  17. CodeQwen2.5-VL: State-of-the-art multimodal LLMgithub.com · 2025-01-30
  18. DiscussionNew Qwen2.5-Max Outperforms DeepSeek V3 in Benchmarkstwitter.com · 2025-01-28
  19. ArticleGuide how to use a local LLM Qwen2.5-coder-7B as an autocompelte in VSCodeeamag.me · 2024-10-29
  20. ArticleQwen2.5-Pluschat.qwenlm.ai · 2025-01-31
  21. ArticleShow HN: I made a free study tool powered by Qwen2.5freeflashcardgenerator.com · 2025-01-28 · 3 HN pts
  22. ArticleQwen2.5 Coderollama.com · 2024-11-11

Where to run it

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

#Cheapest live rowIn /1MOut /1MContextQuantSeen
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; 47 quantized builds for local use; a big base for builders (78 fine-tunes, adapters and merges).

  • As a service1Featherless
  • On own hardware47quantized builds
  • Built on7835 fine-tunes · 6 adapters · 37 merges
  • In apps12Hugging Face Spaces

Top Spaces running it: bethgelab/lm-similarity ♥ 17

Will it fit?

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

Fits a 16 GB card at INT4

About 8.9 GB at INT4, 18 GB at FP8, 35 GB at BF16: weights only, so leave room for the KV cache.

16 GB card64 GB Mac2×80 GB8×80 GBINT4: about 8.9 GB of weights (estimate; KV cache and activations extra)INT48.9 GBFP8: about 18 GB of weights (estimate; KV cache and activations extra)FP818 GBBF16: about 35 GB of weights (estimate; KV cache and activations extra)BF1635 GB
16 GB card64 GB Mac2×80 GB8×80 GBINT4: about 8.9 GB of weights (estimate; KV cache and activations extra)INT48.9 GBFP8: about 18 GB of weights (estimate; KV cache and activations extra)FP818 GBBF16: about 35 GB of weights (estimate; KV cache and activations extra)BF1635 GB

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 (2025-01-23), amber triangles are 4 dated articles (each links to it), grey ticks are repo commits.

OctNovDecJanFebMarAprMayJunJulAugSepOctNovDecJanFebMarAprMayJunJulAugSepOct2025-01-29: 1 repo commit2025-01-28: 2 repo commits2025-01-25: 1 repo commit2025-01-23: 1 repo commitQwen2.5-1M: Deploy your own Qwen with context length up to 1M tokens · 2025-01-26Qwen2.5: A Party of Foundation Models · 2024-09-18Qwen2.5-7B-Instruct-1M and Qwen2.5-14B-Instruct-1M · 2025-01-26Show HN: I made a free study tool powered by Qwen2.5 · 2025-01-28Weights 2025-01-23 · 2025-01-23Weights 2025-01-23now · 1 provider
OctNovDecJanFebMarAprMayJunJulAugSepOctNovDecJanFebMarAprMayJunJulAugSepOct2025-01-29: 1 repo commit2025-01-28: 2 repo commits2025-01-25: 1 repo commit2025-01-23: 1 repo commitQwen2.5-1M: Deploy your own Qwen with context length up to 1M tokens · 2025-01-26Qwen2.5: A Party of Foundation Models · 2024-09-18Qwen2.5-7B-Instruct-1M and Qwen2.5-14B-Instruct-1M · 2025-01-26Show HN: I made a free study tool powered by Qwen2.5 · 2025-01-28Weights 2025-01-23 · 2025-01-23Weights 2025-01-23now · 1 provider

What people built on it

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

125 derivatives on the Hub, led by 47 quantized

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

47 quantized (Hugging Face model tree)35 fine-tunes (Hugging Face model tree)37 merges (Hugging Face model tree)6 adapters (Hugging Face model tree)47 quantized35 fine-tunes37 merges6 adapters
47 quantized (Hugging Face model tree)35 fine-tunes (Hugging Face model tree)37 merges (Hugging Face model tree)6 adapters (Hugging Face model tree)47 quantized35 fine-tunes37 merges6 adapters
GGUF lmstudio-community/Qwen2.5-14B-Instruct-1M-GGUF for llama.cpp, Ollama, LM Studio 107.8k downloads MLX mlx-community/Qwen2.5-14B-Instruct-1M-8bit for Apple silicon 4.3k downloads AWQ graelo/Qwen2.5-14B-Instruct-1M-AWQ for vLLM, SGLang 32.2k downloads

All variants →

Reference

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

What the card says

1 cited sentence

Summary generated, cited

This finetuned model supports a context length of up to 1M tokens.[1]

  1. “supporting a context length of up to 1M tokens”

Generated 2026-10-07 15:09 UTC by parasail-gemma-4-26b-a4b-it; each numeral points at the fact it rests on. Not a verdict.

Use-case evidence

3 use cases

What it is good at 3 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

Pipeline tagtext-generation
Librarytransformers
ArchitecturesQwen2ForCausalLM
Model typeqwen2
Dtypebfloat16
Params by dtypeBF16: 14,770,033,664
Size bandL
MoEno
Context length1,010,000 (config.json max_position_embeddings)
Languagesen
PapersarXiv:2501.15383
LicenseApache 2.0 (apache-2.0)
Hub snapshot2026-10-25

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 Qwen2.5-14B-Instruct-1M: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.

BF16 (2 bytes/param)35 GB · Single high-end GPU (<80 GB)
FP8 (1 byte/param)18 GB · Single GPU (<24 GB)
INT4 (0.5 bytes/param)8.9 GB · Single GPU (<24 GB)

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

All variants

40 variants

Lineage

Fine-tuned from Qwen/Qwen2.5-14B (same organization) Family: Qwen/Qwen2.5 125 derivatives on the Hub

Base model: Qwen/Qwen2.5-14B

quantized: 47finetune: 35merge: 37adapter: 6

Top variants (40 shown, ranked by downloads)

Relation
Quant
40 rows
Variant Relation Quant Params Band License Author Likes Downloads
lmstudio-community/Qwen2.5-14B-Instruct-1M-GGUF quantized GGUF — — apache-2.0 lmstudio-community 24 107.8k
bartowski/Qwen2.5-14B-Instruct-1M-GGUF quantized GGUF — — apache-2.0 bartowski 48 101.1k
DevQuasar-7/Qwen.Qwen2.5-14B-Instruct-1M-GGUF quantized GGUF — — — DevQuasar-7 0 68.5k
mkurman/Qwen2.5-14B-DeepSeek-R1-1M quantized GGUF 14.8B L apache-2.0 mkurman 54 30.5k
graelo/Qwen2.5-14B-Instruct-1M-AWQ quantized AWQ 14.8B L apache-2.0 graelo 2 32.2k
krmonline/Softnix-Qwen2.5-14B-Instruct-1M-LegalThaiCCL-8bit quantized -8bit — — apache-2.0 krmonline 1 25.5k
JungZoona/T3Q-qwen2.5-14b-v1.0-e3 finetune — 14.8B L apache-2.0 JungZoona 59 15.2k
mradermacher/Qwen2.5-14B-Instruct-1M-i1-GGUF quantized GGUF — — apache-2.0 mradermacher 6 11.7k
featherless-ai-quants/Qwen-Qwen2.5-14B-Instruct-1M-GGUF quantized GGUF — — — featherless-ai-quants 0 10.1k
pbhappliedsystems/qwen-2.5-14B-instruct-1m-gguf-Q4-K-M quantized gguf — — apache-2.0 pbhappliedsystems 3 6.6k
pbhappliedsystems/qwen-2.5-14B-instruct-1m-gguf-F16 quantized gguf — — apache-2.0 pbhappliedsystems 0 4.9k
mlx-community/Qwen2.5-14B-Instruct-1M-8bit quantized -8bit 14.8B L apache-2.0 mlx-community 12 4.3k
YOYO-AI/ZYH-LLM-Qwen2.5-14B-V4 merge — 14.8B L apache-2.0 YOYO-AI 8 4.3k
unsloth/Qwen2.5-14B-Instruct-1M-bnb-4bit quantized -bnb 15.2B L apache-2.0 unsloth 2 4.4k
unsloth/Qwen2.5-14B-Instruct-1M finetune — 14.8B L apache-2.0 unsloth 4 3.6k
YOYO-AI/ZYH-LLM-Qwen2.5-14B-V3 merge — 14.8B L apache-2.0 YOYO-AI 8 2.9k
SicariusSicariiStuff/Impish_QWEN_14B-1M finetune — 14.8B L apache-2.0 SicariusSicariiStuff 27 1.9k
unsloth/Qwen2.5-14B-Instruct-1M-unsloth-bnb-4bit quantized -bnb 15.1B L apache-2.0 unsloth 4 3k
mlx-community/Qwen2.5-14B-Instruct-1M-4bit quantized -4bit 14.8B L apache-2.0 mlx-community 2 3k
mradermacher/Qwen2.5-14B-Instruct-1M-GGUF quantized GGUF — — apache-2.0 mradermacher 2 2.9k
tensorblock/Qwen2.5-14B-Instruct-1M-GGUF quantized GGUF — — apache-2.0 tensorblock 0 1.9k
prithivMLmods/Qwen2.5-14B-DeepSeek-R1-1M merge — 14.8B L — prithivMLmods 5 1.7k
krmonline/Softnix-Qwen2.5-14B-Instruct-1M-ThaiFood-Cooking-8bit quantized -8bit — — apache-2.0 krmonline 0 1.9k
prithivMLmods/Calcium-Opus-14B-Elite-1M finetune — 14.8B L apache-2.0 prithivMLmods 4 1.3k
WaveCut/Qwen_Qwen2.5-14B-Instruct-1M_MLX-4bit quantized -4bit 14.8B L apache-2.0 WaveCut 1 1.3k
YOYO-AI/ZYH-LLM-Qwen2.5-14B-V5 merge — 14.8B L apache-2.0 YOYO-AI 8 885
YOYO-AI/Qwen2.5-14B-YOYO-V3 merge — 14.8B L apache-2.0 YOYO-AI 5 833
YOYO-AI/Qwen2.5-14B-YOYO-V4 merge — 14.8B L apache-2.0 YOYO-AI 5 832
mlx-community/Qwen2.5-14B-Instruct-1M-bf16 finetune -bf16 14.8B L apache-2.0 mlx-community 3 849
prithivMLmods/Sombrero-Opus-14B-Sm4 finetune — 14.8B L apache-2.0 prithivMLmods 3 847
mlx-community/Qwen2.5-14B-Instruct-1M-6bit quantized — 14.8B L apache-2.0 mlx-community 2 853
ReadyArt/Omega-Darker_The-Final-Directive-14B finetune — 14.8B L apache-2.0 ReadyArt 9 402
moot20/Qwen2.5-14B-Instruct-1M-MLX-4bit quantized MLX 14.8B L apache-2.0 moot20 0 761
RCorvalan/Qwen2.5-14B-Instruct-1M-Q3_K_M-GGUF quantized GGUF — — apache-2.0 RCorvalan 0 716
YOYO-AI/Qwen2.5-14B-YOYO-V4-p1 merge — 14.8B L apache-2.0 YOYO-AI 0 645
prithivMLmods/Messier-Opus-14B-Sm8 finetune — 14.8B L apache-2.0 prithivMLmods 1 578
ToastyPigeon/Qwen2.5-14B-Instruct-1M-Unalign finetune — 14.8B L — ToastyPigeon 1 552
JamAndTeaStudios/Qwen2.5-14B-Instruct-1M-FP8-Dynamic quantized -FP8 14.8B L apache-2.0 JamAndTeaStudios 0 581
radna/Qwen-1M-DeepSeek-R1-14B merge — 14.8B L — radna 3 431
Lunzima/NQLSG-Qwen2.5-14B-OriginalFusion merge — 14.8B L — Lunzima 2 418

Repository changelog

5 commits

Repository changelog 5 commits

DateCommitSHA
2025-01-29 Update README.md 620fad3
2025-01-28 Update README.md d5c878c
2025-01-28 Update README.md d3a1508
2025-01-25 add README b0c9f6e
2025-01-23 init 261fc41

All providers

1 provider · 1 endpoint

Where to run 1 providers · 1 endpoints

Hosted endpoints observed for Qwen2.5-14B-Instruct-1M, sorted by input price (USD per million tokens). Each row names its observer and date; prices change daily, so check the source before quoting.

ProviderModel refContextQuant$/M in$/M outtok/sTTFTToolsStatusObserved
Featherless
hf
 Qwen/Qwen2.5-14B-Instruct-1M — — — — — — — live 2026-10-25

Articles and coverage

25 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

Qwen2.5-1M: Deploy your own Qwen with context length up to 1M tokens
qwenlm.github.io2025-01-26officialauthorityHN front pageHN 307 pts · 108 comments1,608 words
QwenLM releases open-source Qwen2.5-1M models supporting up to one million token contexts. The 7B and 14B checkpoints significantly outperform their 128K counterparts in long-context tasks while maintaining short-text capabilities. A new vLLM-based inference framework accelerates processing by 3x to 7x faster using Dual Chunk Attention. Deploying the 7B model requires at least 120GB VRAM, while the 14B model needs at least 320GB VRAM.written, reviewed, cited
  • Training begins with a pre-trained checkpoint that originally had a 4K token context length before expansion.
  • The 7B model exhibits minor errors when retrieving information from documents containing up to 1M tokens.
  • Processing 1M-token sequences on the 7B model consumes 71GB of VRAM for MLP layer activations.
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.

Analysis2

Qwen2.5-7B-Instruct-1M and Qwen2.5-14B-Instruct-1M
simonwillison.net2025-01-26authorityHN 98 pts · 2 comments627 words
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 (25) — append-only, nothing is ever removed
TitleSourcePublishedFirst seenSignals
kev: Jev-like model built on Qwen2.5-0.5Bgithub.com2026-09-192026-10-07hn
Civic-SLM is a domain-specialized fine-tune of Qwen2.5-7B for U.S. govt dataitsmeduncan.com2026-04-252026-10-07hn
Show HN: Fine-tuned Qwen2.5-7B on 100 films for probabilistic story graphscinegraphs.ai2026-02-082026-10-07hn
Qwen2.5-Omni Technical Reporthuggingface.co2025-03-302026-10-07hn
Alibaba Qwen2.5-Omni-7B: Open Source End-to-End Multimodal AI Modelalizila.com2025-03-272026-10-07hn
Qwen2.5 Omni: See, Hear, Talk, Write, Do It Allqwenlm.github.io official2025-03-262026-10-07hn
Qwen2.5-VL-32B: Smarter and Lightersimonwillison.net2025-03-252026-10-07hn
Qwen2.5-VL-32B: Smarter and Lighterqwenlm.github.io official2025-03-242026-10-07hn
Qwen2.5-Pluschat.qwenlm.ai2025-01-312026-10-07hn
Qwen2.5-VL: State-of-the-art multimodal LLMgithub.com2025-01-302026-10-07hn
Qwen2.5-Max: The Next Leap in AI Innovationmedium.com2025-01-292026-10-07hn
New Qwen2.5-Max Outperforms DeepSeek V3 in Benchmarkstwitter.com2025-01-282026-10-07hn
Qwen2.5-Max: Exploring the intelligence of large-scale MoE modelqwenlm.github.io official2025-01-282026-10-07hn
Show HN: I made a free study tool powered by Qwen2.5freeflashcardgenerator.com2025-01-282026-10-07hn
Qwen2.5-VLqwenlm.github.io official2025-01-272026-10-07hn
Qwen2.5-7B-Instruct-1M and Qwen2.5-14B-Instruct-1Msimonwillison.net2025-01-262026-10-07hn
Qwen2.5-1M: Deploy your own Qwen with context length up to 1M tokensqwenlm.github.io official2025-01-262026-10-07hn
Qwen2.5-coder on the events occurring during the Tiananmen Square protestsnews.ycombinator.com2024-11-202026-10-07hn
Qwen2.5-Coder-32B is an LLM that can code well that runs on my Macsimonwillison.net2024-11-132026-10-07hn
Qwen2.5-Coder Technical Reportarxiv.org2024-11-112026-10-07hn
Qwen2.5 Coderollama.com2024-11-112026-10-07hn
Qwen2.5-Coder Beats GPT-4o and Claude 3.5 Sonnet in Codingqwenlm.github.io official2024-11-112026-10-07hn
Qwen2.5-Coder-32B with coding abilities matching those of GPT-4ogithub.com2024-11-092026-10-07hn
Guide how to use a local LLM Qwen2.5-coder-7B as an autocompelte in VSCodeeamag.me2024-10-292026-10-07hn
Qwen2.5: A Party of Foundation Modelsqwenlm.github.io official2024-09-182026-10-07hn

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).

Similar models

3 models

Similar models

  • Qwen3-14B · 14.8B
    same size band (L), same task (text-generation), more providers, permissive license
  • phi-4 · 14.7B
    same size band (L), same task (text-generation), more providers, permissive license
  • DeepSeek-R1-Distill-Qwen-14B · 14.8B
    same size band (L), same task (text-generation), more providers, permissive license

Sources

3 sources

Sources

Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:39 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:46 UTC · Summary: 2026-10-07 15:09 UTC · Last verified 2026-10-25

Follow Qwen2.5-14B-Instruct-1M

Watch this model

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