Qwen2.5-72B
Qwen2.5-72B is Alibaba's September 2024 dense LLM snapshot, not the live flagship. Rechecked 2026-08-17: the current Qwen flagship page on this site is Qwen3.8-2.4T-A95B. The 2024 blog still exists at qwen.ai/blog?id=qwen2.5 (the old qwenlm.github.io URL redirects there). Drop leftover flagship / matches Llama-3-405B / pinnacle of the Qwen series as a 2026 claim.
Compare Qwen3.8-2.4T-A95B if you wanted the live MoE flagship, Qwen2.5 72B Instruct if you wanted the instruct twin page, and Qwen Code if you wanted the CLI rather than this 2024 weight.
Key Features
- Previous dense 72B: The 2024 post described 18T pretraining tokens and a 72B dense instruct twin. Recheck Hugging Face for the exact card still hosted.
- Not the 2026 catalog lead: Live ModelStudio rows cited on the Qwen3.8 page are Coder Plus and 397B, not this 72B as flagship.
- No invented API dollar: We did not open a 2026 per-token row for Qwen2.5-72B. Recheck ModelStudio before you quote one.
- Open-weight history, not a bench crown: Do not keep 5x-smaller-than-405B as an audited 2026 ranking.
Use Cases
- People with a 2024 Qwen2.5-72B bookmark who need the catalog corrected.
- People comparing leftover 72B dense vs Qwen3.8 MoE.
- People who were about to paste Llama-3-405B parity into a 2026 deck.
Limitation: We did not re-download the 2024 weight card or re-run MMLU/HumanEval.
Pricing
First-party pages on 2026-08-17.
| Piece | Price | Notes |
|---|---|---|
| Qwen2.5-72B weights | Previous snapshot | Recheck Hugging Face / ModelStudio. Not the live flagship. |
| Live Qwen flagship | Recheck Qwen3.8 page | Coder Plus $0.14/$0.80 and 397B $0.29/$1.20 were checked there. |
If a leftover blog still calls 72B the Qwen flagship, treat this page as the correction.
Getting Started
- Open qwen.ai/blog?id=qwen2.5 only for the 2024 note.
- Recheck Qwen3.8-2.4T-A95B before you buy a 2026 seat.
- Recheck Qwen Code if you wanted the CLI.
- Do not paste Llama-3-405B parity into a deck.
Frequently Asked Questions
Is Qwen2.5-72B still the flagship?
No. The live flagship page on this site is Qwen3.8.
Did it officially match Llama-3-405B?
That leftover 2024 claim is out as a 2026 crown.
Same as Qwen2.5 72B Instruct?
Related twin page. This slug is the 72B family snapshot.
Alternatives
- Qwen3.8-2.4T-A95B: Live Qwen flagship.
- Qwen Code: Alibaba open CLI.
- GPT-5.6 Sol: OpenAI live flagship if you wanted a 2026 frontier chat model.
Tips
- Call 72B a September 2024 snapshot.
- Point new work at Qwen3.8.
- Recheck ModelStudio before you invent a 72B token price.
Conclusion
Qwen2.5-72B is a 2024 dense snapshot, not the 2026 Qwen flagship and not a Llama-3-405B crown. Start at the Qwen3.8 page, then decide whether that live SKU already covers the model you need.
Comments
No comments yet. Be the first to comment!
Related Tools
Qwen2.5 72B Instruct
qwen.ai/blog?id=qwen2.5
Previous Qwen2.5 72B Instruct snapshot. Rechecked: live flagship is Qwen3.8. Not a 2026 latest series.
Qwen2.5-Coder-32B
qwen.ai/blog?id=qwen2.5-coder-family
Previous Qwen code snapshot. Rechecked: live coding path is Qwen3.8 Coder. Not an 85% HumanEval census.
Shap-e
github.com/openai/shap-e
Previous OpenAI Shap-E snapshot. Rechecked: leftover 2023 3D research card, not a 2026 image/3D default. Live stills path is GPT Image 2.
Related Insights
Grok Bot and Hermes Bot: one person finally gets a think tank and a secretariat
Grok Bot now ships with Cursor Pro+. Hermes Bot runs on a VPS. They are not smarter chat boxes. The think tank advises, the secretariat executes, and you still make the call.
Hook OpenCode Go into Codex on Windows. Do Not Open a Second Toolkit.
A ChatGPT-signed Codex desktop app still shows mostly GPT in the picker. On Windows, enable only OpenCode Go and the Grok, GLM, Kimi, DeepSeek, and MiniMax models you already pay for appear in the same selector. Keys stay local. Native GPT stays put.
What Locks Codex Is the Picker, Not the Models
You already pay for OpenCode Go, Grok, and Z.ai, but the Codex picker still shows mostly GPT. The community project codex-router does not teach another install ritual. It puts subscriptions you already bought back into the selector, keeps keys on the machine, and leaves native GPT alone.