GPT OSS 120B vs GPT OSS 20B vs Qwen3 235B-A22B Instruct 2507
Too close to call on our weighted score (GPT OSS 120B 57, Qwen3 235B-A22B Instruct 2507 55, GPT OSS 20B 53). The right pick depends on what you value most.
Price in rating: Excluded · Include price
OpenAI
GPT OSS 120B
57/100- ECI140.2
- Price$0.15 / $0.60
- Context131K
OpenAI
GPT OSS 20B
53/100- ECI134.9
- Price$0.05 / $0.20
- Context131K
Alibaba (Qwen)
Qwen3 235B-A22B Instruct 2507
55/100- ECI138.1
- Price$0.15 / $0.60
- Context262K
Too close to call
It is close. Our weighted score puts them within 2 points (GPT OSS 120B 57/100, Qwen3 235B-A22B Instruct 2507 55/100, GPT OSS 20B 53/100), so choose by what matters most for your work: GPT OSS 120B for raw capability, GPT OSS 20B on price and Qwen3 235B-A22B Instruct 2507 for long inputs. The score weighs capability 67%, inputs & features 20%, context window 13%.
- CapabilityGPT OSS 120BCapabilities Index (ECI): GPT OSS 120B 140.2 · Qwen3 235B-A22B Instruct 2507 138.1 · GPT OSS 20B 134.9
- Lowest priceGPT OSS 20BGPT OSS 20B $0.087 · GPT OSS 120B $0.263 · Qwen3 235B-A22B Instruct 2507 $0.263 per 1M tokens (3:1 blend)
- Longest contextQwen3 235B-A22B Instruct 2507Qwen3 235B-A22B Instruct 2507 262,144 · GPT OSS 120B 131,072 · GPT OSS 20B 131,072 tokens
- Widest inputsSame inputsGPT OSS 120B: Text · GPT OSS 20B: Text · Qwen3 235B-A22B Instruct 2507: Text
- Self-hostingAll open weightsEvery model here can be downloaded and run on your own hardware
| Measure | Weight | GPT OSS 120B | GPT OSS 20B | Qwen3 235B-A22B Instruct 2507 |
|---|---|---|---|---|
| CapabilityCapabilities Index (ECI) | 67% | 68 | 63 | 66 |
| Inputs & features | 20% | 35 | 35 | 25 |
| Context window | 13% | 32 | 32 | 43 |
| Overall | 100% | 57/100 | 53/100 | 55/100 |
Every spec in one table
Highlighted cells lead their row. Dashes mean the data is not published.
| Specification | |||
|---|---|---|---|
| Capability | |||
| Capabilities Index (ECI) | 140.2 (best) | 134.9 | 138.1 |
| ECI rank | #115 of 158 (best) | #126 of 158 | #120 of 158 |
| GPQA DiamondGraduate-level science questions | 75.8% (best) | 60.8% | — |
| OTIS Mock AIME 2024–2025Competition mathematics | 88.9% (best) | 65.3% | — |
| Price per million tokens | |||
| Input | $0.15 | $0.05 (best) | $0.15 |
| Output | $0.60 | $0.20 (best) | $0.60 |
| Cached input | — | — | — |
| Blended (3:1) | $0.263 | $0.087 (best) | $0.263 |
| Long-context rate | Same rate | Same rate | Same rate |
| Price source | Median of 77 providers | Median of 39 providers | Median of 23 providers |
| Limits | |||
| Context window | 131,072 tokens | 131,072 tokens | 262,144 tokens (best) |
| Max output | 131,072 tokens (best) | 131,072 tokens (best) | 16,384 tokens |
| Inputs and features | |||
| Text | Yes | Yes | Yes |
| Images | No | No | No |
| PDFs | No | No | No |
| Audio | No | No | No |
| Video | No | No | No |
| Reasoning | Yes | Yes | No |
| Tool calling | Yes | Yes | Yes |
| Structured output | No | No | No |
| Availability | |||
| Weights | Open | Open | OpenApache 2.0 |
| API model ID | — | — | — |
| API providers | 80 (best) | 44 | 25 |
| Released | Aug 5, 2025 | Aug 5, 2025 | Jul 21, 2025 |
| Knowledge cutoff | Jun 2024 | Jun 2024 | — |
What would a month cost?
Enter your expected volume in millions of tokens. List prices only; caching and batch discounts would lower these.
GPT OSS 120B$2.70
GPT OSS 20B$0.90
Qwen3 235B-A22B Instruct 2507$2.70
Which should you choose?
Which is better: GPT OSS 120B, GPT OSS 20B or Qwen3 235B-A22B Instruct 2507?
It is close. Our weighted score puts them within 2 points (GPT OSS 120B 57/100, Qwen3 235B-A22B Instruct 2507 55/100, GPT OSS 20B 53/100), so choose by what matters most for your work: GPT OSS 120B for raw capability, GPT OSS 20B on price and Qwen3 235B-A22B Instruct 2507 for long inputs. The score weighs capability 67%, inputs & features 20%, context window 13%.
Which is cheaper, GPT OSS 120B, GPT OSS 20B or Qwen3 235B-A22B Instruct 2507?
GPT OSS 20B is cheaper at $0.05 input / $0.20 output per million tokens (median across 39 API providers). GPT OSS 120B costs $0.15 input / $0.60 output per million tokens (median across 77 API providers); Qwen3 235B-A22B Instruct 2507 costs $0.15 input / $0.60 output per million tokens (median across 23 API providers). At a typical mix of three input tokens to one output token, that is $0.087 per million tokens for GPT OSS 20B versus $0.263 for GPT OSS 120B (3× as much) and $0.263 for Qwen3 235B-A22B Instruct 2507 (3× as much).
Which scores higher on benchmarks?
GPT OSS 120B scores higher on the Capabilities Index (ECI): GPT OSS 120B 140.2 (#115 of 158), Qwen3 235B-A22B Instruct 2507 138.1 (#120 of 158) and GPT OSS 20B 134.9 (#126 of 158). The confidence ranges of the top two overlap (138.3–142.1 vs 135.7–140.2), so treat the gap as small.
Which is better for coding?
There are no published SWE-bench Verified results for GPT OSS 120B, GPT OSS 20B and Qwen3 235B-A22B Instruct 2507 yet, so there is no like-for-like coding score. On overall capability, GPT OSS 120B leads, which tends to carry over to coding, but test on your own codebase. All three support tool calling for agent workflows.
Which has the bigger context window?
Qwen3 235B-A22B Instruct 2507 has the largest context window at 262,144 tokens, against 131,072 for GPT OSS 120B and 131,072 for GPT OSS 20B. Maximum output per response: GPT OSS 120B up to 131,072, GPT OSS 20B up to 131,072, Qwen3 235B-A22B Instruct 2507 up to 16,384 tokens.
Which can read images, PDFs, audio or video?
GPT OSS 120B accepts text; GPT OSS 20B accepts text; Qwen3 235B-A22B Instruct 2507 accepts text. They handle the same number of input types.
Are any of these open source?
Yes, all three publish their weights (Apache 2.0), so you can self-host them.
Which is newer?
GPT OSS 120B is the newest, released Aug 5, 2025. GPT OSS 20B came out Aug 5, 2025; Qwen3 235B-A22B Instruct 2507 came out Jul 21, 2025. Knowledge cutoff: GPT OSS 120B Jun 2024, GPT OSS 20B Jun 2024.
How do you decide the winner?
This rating uses capability (67%, independent benchmark results); inputs & features (20%, image, PDF, audio and video input, tool calling, structured output and reasoning); context window (13%, maximum tokens per request, log scale). Dimensions missing for any model are dropped and the remaining weights rescaled, so every model is judged on the same evidence. Specs and prices come from public model listings and the labs’ own API pages; capability scores come from independent benchmark runs. Data updated Nov 18, 2026.