gpt-oss-120b vs Gemma 4 31B
A side-by-side look at gpt-oss-120b from OpenAI and Gemma 4 31B from Google: list prices, context windows, licensing and capabilities, with monthly costs worked out for three common workloads.
The short version
- Price: effectively tied on blended list price.
- Larger context window: Gemma 4 31B (256K vs 128K tokens).
- Open weights: both can be self-hosted (Apache-2.0 and Apache-2.0).
- Inputs: gpt-oss-120b accepts text; Gemma 4 31B accepts text, images, video.
- Newer release: Gemma 4 31B (March 31, 2026).
Specs side by side
| gpt-oss-120b | Gemma 4 31B | |
|---|---|---|
| Provider | OpenAI | |
| Released | Aug 5, 2025 | Mar 31, 2026 |
| Status | Current | Current |
| Weights and license | Open weights Apache-2.0 | Open weights Apache-2.0 |
| Tier | Advanced | Efficient |
| Context window | 128K | 256K |
| Max output | — | — |
| Input per 1M tokens | $0.15 | $0.20 |
| Output per 1M tokens | $0.60 | $0.40 |
| Cached input per 1M | $0.015 | — |
| Input types | Text | Text, Images, Video |
| Reasoning mode | Yes | Yes |
| Tool calling | Yes | Yes |
| Parameters | 117B total, 5.1B active | 31.3B |
Highlighted cells are better on that line: lower price or larger limit. List prices, standard tier, checked October 10, 2026.
What they cost for real workloads
| Workload (monthly) | gpt-oss-120b | Gemma 4 31B | Difference |
|---|---|---|---|
| Support chatbot30,000 conversations a month, 2,000 input and 400 output tokens each | $16 | $17 | gpt-oss-120b saves $0.60 |
| Document summaries2,000 documents a month, 12,000 input and 800 output tokens each | $4.56 | $5.44 | gpt-oss-120b saves $0.88 |
| Coding agent1,000 tasks a month, 60,000 input and 6,000 output tokens each | $13 | $14 | gpt-oss-120b saves $1.80 |
Standard rates without caching or batch discounts. Try your own volumes.
Choose gpt-oss-120b if
- you mainly need it for reasoning, agents, high volume, which OpenAI emphasizes for this model
- fits on one 80 GB GPU thanks to MXFP4 weights
- adjustable reasoning effort with browsing, Python and function-calling formats
Choose Gemma 4 31B if
- you mainly need it for multimodal, multilingual, fine-tuning, which Google emphasizes for this model
- you need more than 128K tokens of context in one request
- your inputs include images, video
- apache 2.0 license (earlier Gemma versions used custom terms)
- configurable thinking mode and native function calling
Our take
Both are positioned for self-hosting. On list price the gap is about 1.0×, which matters most for high-volume, input-heavy workloads. Quality differences shift with every release, so treat this page as a shortlist tool: run 20-50 of your real tasks through both, score the outputs, and divide the bill by the number of tasks each one got right.
Read the full profiles: gpt-oss-120b and Gemma 4 31B. For help designing that test, see how to read LLM benchmarks.
Frequently asked questions
Which is cheaper, gpt-oss-120b or Gemma 4 31B?
Gemma 4 31B has the lower list price: $0.20 input and $0.40 output per million tokens, versus $0.15 and $0.60. Discounts for caching and batch processing can change the real bill, so price your own token mix.
Which has the bigger context window?
Gemma 4 31B accepts 256K tokens, compared with 128K for gpt-oss-120b.
Can I run gpt-oss-120b or Gemma 4 31B on my own hardware?
gpt-oss-120b is open-weight (Apache-2.0), so yes, with enough GPU memory. Gemma 4 31B is open-weight (Apache-2.0), so yes, with enough GPU memory.
Is gpt-oss-120b better than Gemma 4 31B?
It depends on the task. Public benchmarks and arena rankings move with every release and rarely match a specific workload. Use the specs here to shortlist, then test both on your own prompts and compare accuracy, latency and cost per correct answer.
Still choosing a model?
Send us your use case and constraints. We'll come back with a shortlist and an evaluation plan you can run on your own data.
- Free and vendor-neutral: no model provider pays for placement.
- Your details go only to partners you agree to.