// head_to_head
Gemma 4 31B vs Qwen2.5 72B Instruct
Published LiveBench scores across all seven categories, live list pricing, context windows, and the measured cost of a point of capability — for both models, side by side.
Gemma 4 31B is the cheaper of the two; neither can be ranked on quality here.
Qwen2.5 72B Instruct does not have a published LiveBench run, so this comparison covers price, context and declared capabilities only. A missing score means "not evaluated", not "worse" — the right way to separate these two is an eval on your own workload.
Gemma 4 31B
- Blended / 1M
- $0.250
- Context
- 262K
- Released
- Apr 2, 2026
- Overall score
- 38.7
qwen
Qwen2.5 72B Instruct
- Blended / 1M
- $0.370
- Context
- 33K
- Released
- Sep 19, 2024
- Overall score
- Not evaluated
Specs and pricing
| Metric | Gemma 4 31B | Qwen2.5 72B Instruct |
|---|---|---|
| LiveBench overall Weighted mean of the seven LiveBench category scores, 0–100. Higher is better. | 38.7 | — |
| Cost per point Measured benchmark spend divided by overall score — dollars per point of capability. | $0.0017 | — |
| Blended price / 1M 3:1 input:output mix, the usual shape of production traffic. | $0.250win | $0.370 |
| Input price / 1M | $0.200win | $0.360 |
| Output price / 1M | $0.400 | $0.400 |
| Cached input / 1M Price of an input token served from the prompt cache, where the provider publishes one. | — | — |
| Context window | 262Kwin | 33K |
| Max output tokens | 16K | 16K |
Benchmarks by category
An overall score averages away the thing you probably care about. These are the seven LiveBench categories scored separately — Gemma 4 31B on top, Qwen2.5 72B Instruct below, both out of 100.
What each one costs to run
Per-token prices are hard to feel. These are monthly list costs for both models across five workload shapes, using each provider's published cached-input rate where there is one.
| Workload | Gemma 4 31B | Qwen2.5 72B Instruct |
|---|---|---|
| Support chatbot 1.2K in / 400 out × 200K requests | $80.00/mo | $118.40/mo |
| RAG assistant 8K in / 600 out × 100K requests | $184.00/mo | $312.00/mo |
| Coding agent 40K in / 4K out × 20K requests | $192.00/mo | $320.00/mo |
| Document extraction 20K in / 1.5K out × 50K requests | $230.00/mo | $390.00/mo |
| Bulk classification 500 in / 20 out × 5M requests | $540.00/mo | $940.00/mo |
Which should you pick?
You need to fit large documents in one call
Gemma 4 31B
Wider context window — 262K against 33K.
Gemma 4 31B vs Qwen2.5 72B Instruct FAQ
Which is better, Gemma 4 31B or Qwen2.5 72B Instruct?
Gemma 4 31B is the cheaper of the two; neither can be ranked on quality here. Qwen2.5 72B Instruct does not have a published LiveBench run, so this comparison covers price, context and declared capabilities only. A missing score means "not evaluated", not "worse" — the right way to separate these two is an eval on your own workload.
Is Gemma 4 31B cheaper than Qwen2.5 72B Instruct?
Gemma 4 31B is cheaper. On a 3:1 input:output blend, Gemma 4 31B lists at $0.250 per million tokens and Qwen2.5 72B Instruct at $0.370 — Gemma 4 31B is 48% cheaper. Input and output are priced separately — Gemma 4 31B charges $0.200 in and $0.400 out, Qwen2.5 72B Instruct charges $0.360 and $0.400 — so the model that looks cheaper flips depending on how output-heavy your workload is.
Does Gemma 4 31B or Qwen2.5 72B Instruct have a bigger context window?
Gemma 4 31B has the larger context window: 262K for Gemma 4 31B against 33K for Qwen2.5 72B Instruct. Note that a window you can fill is not a window you should fill — retrieval quality usually degrades well before the limit, and you pay for every token you put in it.
Related comparisons
How these numbers are produced
- Price — provider list price from OpenRouter, refreshed every 15 minutes. “Blended” is a 3:1 input:output mix.
- Scores — LiveBench release 2026-10-25, using their own category map. Each model shows its strongest published run. A blank means “not evaluated”, never “bad”.
- Cost per point — the measured dollars LiveBench spent on the run, divided by the score it earned.
- “Win” — awarded only past a threshold: one full point on a benchmark score, 10% on a price, 25% on a context window. Anything tighter reports as a tie, because effort settings alone move a LiveBench score by more than that.
Published benchmarks rank models on someone else's tasks. Before committing, see LLM & agent evaluation for building an eval on your own.