Specification

Family: Qwen
Source:Qwen — official blogT2observed 9 h agomediumLLM-extracted
Version: 2.5
Source:Qwen — official blogT2observed 9 h agomediumLLM-extracted
Release date: 19 Sept 2024
Source:Qwen — official blogT2observed 9 h agomediumLLM-extracted
Status: active
Source:Qwen — official blogT2observed 9 h agomediumLLM-extracted
Openness: open-weights
Source:Qwen — official blogT2observed 9 h agomediumLLM-extracted
License: Apache 2.0 (most models), Qwen Research License (Qwen2.5-3B), Qwen License (Qwen2.5-72B)
Source:Qwen — official blogT2observed 9 h agomediumLLM-extracted
Architecture: decoder-only dense
Source:Qwen — official blogT2observed 9 h agomediumLLM-extracted
Parameters: 72.7B
Source:Qwen — official blogT2observed 9 h agomediumLLM-extracted
Active parameters: 70B
Source:Qwen — official blogT2observed 9 h agomediumLLM-extracted
Mixture of experts: No
Source:Qwen — official blogT2observed 9 h agomediumLLM-extracted
Context window: 131.1K tokens
Source:Qwen — official blogT2observed 9 h agomediumLLM-extracted
Max output: 8.19K tokens
Source:Qwen — official blogT2observed 9 h agomediumLLM-extracted
Modalities: text
Source:Qwen — official blogT2observed 9 h agomediumLLM-extracted
Input modalities: text
Source:Qwen — official blogT2observed 9 h agomediumLLM-extracted
Output modalities: text
Source:Qwen — official blogT2observed 9 h agomediumLLM-extracted
Languages: Arabic, English, French, Indonesian, Japanese, Korean, Portuguese, Spanish, Turkish, Vietnamese
Source:Qwen — official blogT2observed 9 h agomediumLLM-extracted
Official page: qwen.ai
Source:Qwen — official blogT2observed 9 h agomediumLLM-extracted
Paper: qwenlm.github.io/blog/qwen2.5
Source:Qwen — official blogT2observed 9 h agomediumLLM-extracted
Repository: huggingface.co/Qwen
Source:Qwen — official blogT2observed 9 h agomediumLLM-extracted

Each value shows its source, tier and observation time. Conflicting claims are kept side by side and flagged — never averaged. How AI Atlas records facts →

Provenance

Attributed facts

26

Source tiers

T226

Freshest observation

9 h ago

Conflicts

None

Modalities

Modalities: text
Input: text
Output: text

Capabilities

Tool calling
Yes
Qwen — official blog · T2
Structured output
Yes
Qwen — official blog · T2
Reasoning
Yes
Qwen — official blog · T2
Vision
No
Qwen — official blog · T2
Audio
No
Qwen — official blog · T2
Fine-tuning available
Yes
Qwen — official blog · T2

Context window: 131.1K tokens
Source:Qwen — official blogT2observed 9 h agomediumLLM-extracted
Max output: 8.19K tokens
Source:Qwen — official blogT2observed 9 h agomediumLLM-extracted
Languages: Arabic, English, French, Indonesian, Japanese, Korean, Portuguese, Spanish, Turkish, Vietnamese
Source:Qwen — official blogT2observed 9 h agomediumLLM-extracted

No benchmark results recorded

Results appear when a tier 1–3 source publishes them; we never copy scores without a source.

Current prices

No current prices recorded

Prices appear when a provider publishes a public pricing page we crawl.

Price history

Estimated

Memory need = bytes per parameter (4-bit ≈ 0.5 × 1.15 overhead, 8-bit 1.0, fp16 2.0) + a KV-cache allowance. Not a measurement.

Estimated hardware fit
Hardware	Quantization	Memory	Est. need	Fits
NVIDIA DGX B200	4bit	1,440 GB	42.3 GB	Yes
Apple M3 Ultra	4bit	—	42.3 GB	Yes
Mac Studio (Apple M5 Ultra)	4bit	—	42.3 GB	Yes
AMD Instinct MI325X	4bit	256 GB	42.3 GB	Yes
AMD Instinct MI300X	4bit	192 GB	42.3 GB	Yes
Apple M2 Ultra	4bit	—	42.3 GB	Yes
NVIDIA B200	4bit	180 GB	42.3 GB	Yes
NVIDIA H200	4bit	141 GB	42.3 GB	Yes
NVIDIA H200 NVL	4bit	141 GB	42.3 GB	Yes
Apple M1 Ultra	4bit	—	42.3 GB	Yes
Apple M3 Max	4bit	—	42.3 GB	Yes
Apple M4 Max	4bit	—	42.3 GB	Yes
Mac Studio (Apple M5 Max)	4bit	—	42.3 GB	Yes
MacBook Pro (Apple M5 Max)	4bit	—	42.3 GB	Yes
NVIDIA DGX Spark	4bit	128 GB	42.3 GB	Yes
Apple M2 Max	4bit	—	42.3 GB	Yes
NVIDIA H100 NVL	4bit	94 GB	42.3 GB	Yes
NVIDIA A100 80GB	4bit	80 GB	42.3 GB	Yes
NVIDIA H100 SXM	4bit	80 GB	42.3 GB	Yes
Apple M1 Max	4bit	—	42.3 GB	Yes
Apple M4 Pro	4bit	—	42.3 GB	Yes
Mac mini (Apple M5 Pro)	4bit	—	42.3 GB	Yes
MacBook Pro (Apple M5 Pro)	4bit	—	42.3 GB	Yes
Apple M3 Pro	4bit	—	42.3 GB	No
Apple M1 Pro	4bit	—	42.3 GB	No
Apple M2 Pro	4bit	—	42.3 GB	No
Apple M4	4bit	—	42.3 GB	No
iMac (Apple M4)	4bit	—	42.3 GB	No
Mac mini (Apple M6)	4bit	—	42.3 GB	No
MacBook Air (Apple M5)	4bit	—	42.3 GB	No
MacBook Pro (Apple M5)	4bit	—	42.3 GB	No
NVIDIA GeForce RTX 5090	4bit	32 GB	42.3 GB	No
Apple M2	4bit	—	42.3 GB	No
Apple M3	4bit	—	42.3 GB	No
NVIDIA GeForce RTX 3090	4bit	24 GB	42.3 GB	No
NVIDIA GeForce RTX 4090	4bit	24 GB	42.3 GB	No
Apple M1	4bit	—	42.3 GB	No

Ancestors 1

Qwen2

This model

Qwen2.5

72.7B params

Quantizations 0

None recorded.

Descendants 2

Qwen2.5-Coder32.5B
Qwen2.5-MathQwen Team

Papers 0

No papers linked yet.

Repositories 0

No repositories linked yet.

As of

Rewind the record: see this entity's attributes exactly as AI Atlas knew them on a given day.

Claim history · Training data notes

1 claims · 1 propertiesShow all properties

Training data notestraining_data_notes1

Claim history for Training data notes
Value	Valid from → to	Status	Source	Confidence	Extractor
Pretrained on up to 18 trillion tokens. Qwen2.5-Coder trained on 5.5 trillion tokens of code-related data. Qwen2.5-Math pretrained on larger-scale math related data including synthetic data generated by Qwen2-Math.	11 Sept 2026 20:59 UTC → current	current	Qwen — official blogT2	medium	llm

Claims are temporal and append-only: a new observation closes the previous claim (valid_to) instead of overwriting it. Conflicting claims from different sources are kept side by side and flagged — never averaged. Methodology →

11 Sept 2026
Property changedModelQwen2.5Qwen Team
Qwen2.5: official url changed from https://qwen.ai to qwen.ai
Official pagehttps://qwen.ai→qwen.aiqwen
qwenlm.github.io
11 Sept 2026
Property changedModelQwen2.5Qwen Team
Qwen2.5: languages changed from ["Arabic", "Chinese", "English", "French", "German", "Ita… to ["Arabic", "English", "French", "Indonesian", "Japanese",…
LanguagesArabic, Chinese, English, French, German, Italian, Japanese, Korean, Portuguese, Russian, Spanish, Thai, Vietnamese→Arabic, English, French, Indonesian, Japanese, Korean, Portuguese, Spanish, Turkish, Vietnameseqwen
qwenlm.github.io
11 Sept 2026
Max output changedModelQwen2.5Qwen Team
Qwen2.5: max output tokens changed from 8000 to 8192
Max output8K tokens→8.19K tokensqwen
qwenlm.github.io
11 Sept 2026
Context window changedModelQwen2.5Qwen Team
Qwen2.5: context length changed from 128000 to 131072
Context window128K tokens→131.1K tokensqwen
qwenlm.github.io
11 Sept 2026
Parameters changedModelQwen2.5Qwen Team
Qwen2.5: parameter count changed from 72000000000 to 72700000000
Parameters72B→72.7Bqwen
qwenlm.github.io
11 Sept 2026
Property changedModelQwen2.5Qwen Team
Qwen2.5: architecture changed from dense, decoder-only to decoder-only dense
Architecturedense, decoder-only→decoder-only denseqwen
qwenlm.github.io
11 Sept 2026
License changedModelQwen2.5Qwen Team
Qwen2.5: license changed from Apache 2.0 to Apache 2.0 (most models), Qwen Research License (Qwen2.5-…
LicenseApache 2.0→Apache 2.0 (most models), Qwen Research License (Qwen2.5-3B), Qwen License (Qwen2.5-72B)qwen
qwenlm.github.io
19 Sept 2024
New modelModelQwen2.5Qwen Team
New model: Qwen2.5 (Qwen Team)
qwen
qwenlm.github.io

Full timeline →

Source documents
Source	Document	Type	Tier	Last observed	Snapshots
Qwen — official blog	qwenlm.github.io/blog/qwen2.5-math	news	T1· Official	9 h ago	1
Qwen — official blog	qwenlm.github.io/blog/qwen2.5-coder	news	T1· Official	9 h ago	1
Qwen — official blog	qwenlm.github.io/blog/qwen2.5-llm	news	T1· Official	9 h ago	1
Qwen — official blog	qwenlm.github.io/blog/qwen2.5	news	T1· Official	9 h ago	1
Qwen — official blog	qwenlm.github.io/blog/qwen2.5-turbo	news	T1· Official	9 h ago	1
Qwen — official blog	qwenlm.github.io/blog/qwen3	news	T1· Official	9 h ago	1

Tier 1 = official/primary, 2 = quality secondary, 3 = community, 4 = unverified. Every snapshot is archived; see all sources and the methodology.