Qwen2.5 72B Instruct

Qwen (Alibaba) · Budget · Context 131K

qwen/qwen-2.5-72b-instruct

Data as of:

LLM API list prices change frequently (new models and price cuts are common) and vary by tier, region, batch / cache usage and time. These are list prices captured at the time shown; always verify the current price with the provider before relying on it.

Price summary

Input $/1M $0.36

per 1M input tokens

Output $/1M $0.4

per 1M output tokens

Blended $/1M $0.37

0.75×input + 0.25×output (factual)

Blended $/1M is a published convenience figure: 0.75 × input + 0.25 × output (a stated 3:1 input:output mix). It is descriptive arithmetic, not a value verdict.

Specifications

Model
Qwen2.5 72B Instruct
Provider
Qwen (Alibaba)
Input $/1M
$0.36
Output $/1M
$0.4
In+Out $/1M
$0.76
Context
131K tokens
Max output
16K tokens
Cache read $/1M
Modalities
text → text
Cross-checked
Differs

Capability

Capability score
48.0
MMLU-PRO
51.4
GPQA
16.7

Capability values are the published per-model score from Open LLM Leaderboard (Hugging Face), shown as-is with no edit and no “best” verdict. The leaderboard evaluates open-weight models only and lags the newest releases, so many models (including closed/proprietary APIs) have no value and show “—”. Different benchmarks rank models differently; treat this as one signal among many. As of 2026-05-25. Open LLM Leaderboard (Hugging Face) (Apache-2.0).

Official benchmark (maker-published)

MMLU (official)
86.1% (5-shot)
GPQA-Diamond (official)

These are the model maker's own published benchmark scores, reproduced as-is with the publisher source and an as-of date — not a Quanteta score and not a recommendation. They are raw percentages on the named benchmark and are NOT on the same scale as the open-weight leaderboard scores above; do not compare the two directly. The exact evaluation setting (e.g. 5-shot vs 0-shot chain-of-thought) is shown per value because it changes the number; only same-setting values are plotted together. Source: Qwen2.5 Technical Report (arXiv:2412.15115) / Qwen2.5-LLM blog: Qwen2.5-72B MMLU 86.1 (5-shot) (as of 2024-12-20).

Try it / official references

External links open the provider's own pages; list prices and availability there are authoritative.

Estimated cost per use case

Use caseinput tokensoutput tokensCost (per 1,000 requests)
Chat / assistant 1,000 500 $0.56
RAG / Q&A 8,000 800 $3.2
Coding agent 6,000 2,000 $2.96
Summarization 12,000 600 $4.56

Each row is (input_tokens/1M)×input_price + (output_tokens/1M)×output_price, scaled to 1,000 requests. Assumptions are as shown in the table. Not a recommendation.