GPT OSS 120B vs GPT-5.4 nano vs Qwen3 235B-A22B Instruct 2507
GPT-5.4 nano comes out ahead, 68 to 61 and 60 on our weighted score, though GPT OSS 120B is 43% cheaper per token.
Price in rating: Included · Exclude price
OpenAI
GPT OSS 120B
61/100- ECI140.2
- Price$0.15 / $0.60
- Context131K
- Our pick
OpenAI
GPT-5.4 nano
68/100- ECI146.3
- Price$0.20 / $1.25
- Context400K
Alibaba (Qwen)
Qwen3 235B-A22B Instruct 2507
60/100- ECI138.1
- Price$0.15 / $0.60
- Context262K
GPT-5.4 nano is our pick
GPT-5.4 nano is the better all-round choice, scoring 68/100 against GPT OSS 120B (61) and Qwen3 235B-A22B Instruct 2507 (60). It leads on capability, inputs & features and context window. The score weighs capability 50%, price 25%, inputs & features 15%, context window 10%.
- CapabilityGPT-5.4 nanoCapabilities Index (ECI): GPT-5.4 nano 146.3 · GPT OSS 120B 140.2 · Qwen3 235B-A22B Instruct 2507 138.1
- Lowest priceGPT OSS 120B and Qwen3 235B-A22B Instruct 2507GPT OSS 120B $0.263 · Qwen3 235B-A22B Instruct 2507 $0.263 · GPT-5.4 nano $0.463 per 1M tokens (3:1 blend)
- Longest contextGPT-5.4 nanoGPT-5.4 nano 400,000 · Qwen3 235B-A22B Instruct 2507 262,144 · GPT OSS 120B 131,072 tokens
- Widest inputsGPT-5.4 nanoGPT OSS 120B: Text · GPT-5.4 nano: Text, Images · Qwen3 235B-A22B Instruct 2507: Text
- Self-hostingGPT OSS 120B and Qwen3 235B-A22B Instruct 2507Publishes downloadable weights (Apache 2.0)
| Measure | Weight | GPT OSS 120B | GPT-5.4 nano | Qwen3 235B-A22B Instruct 2507 |
|---|---|---|---|---|
| CapabilityCapabilities Index (ECI) | 50% | 68 | 74 | 66 |
| Price | 25% | 74 | 63 | 74 |
| Inputs & features | 15% | 35 | 70 | 25 |
| Context window | 10% | 32 | 50 | 43 |
| Overall | 100% | 61/100 | 68/100 | 60/100 |
Every spec in one table
Highlighted cells lead their row. Dashes mean the data is not published.
| Specification | |||
|---|---|---|---|
| Capability | |||
| Capabilities Index (ECI) | 140.2 | 146.3 (best) | 138.1 |
| ECI rank | #115 of 158 | #86 of 158 (best) | #120 of 158 |
| GPQA DiamondGraduate-level science questions | 75.8% | 78.5% (best) | — |
| FrontierMath Tiers 1–3Research-level mathematics | — | 44.9% | — |
| OTIS Mock AIME 2024–2025Competition mathematics | 88.9% (best) | 87.8% | — |
| SimpleQA VerifiedShort factual questions | — | 11.7% | — |
| Price per million tokens | |||
| Input | $0.15 (best) | $0.20 | $0.15 (best) |
| Output | $0.60 (best) | $1.25 | $0.60 (best) |
| Cached input | — | $0.02 | — |
| Blended (3:1) | $0.263 (best) | $0.463 | $0.263 (best) |
| Long-context rate | Same rate | Same rate | Same rate |
| Price source | Median of 77 providers | Official OpenAI API | Median of 23 providers |
| Limits | |||
| Context window | 131,072 tokens | 400,000 tokens (best) | 262,144 tokens |
| Max output | 131,072 tokens (best) | 128,000 tokens | 16,384 tokens |
| Inputs and features | |||
| Text | Yes | Yes | Yes |
| Images | No | Yes | No |
| PDFs | No | No | No |
| Audio | No | No | No |
| Video | No | No | No |
| Reasoning | Yes | Yeslow · medium · high · xhigh | No |
| Tool calling | Yes | Yes | Yes |
| Structured output | No | Yes | No |
| Availability | |||
| Weights | Open | Proprietary | OpenApache 2.0 |
| API model ID | — | gpt-5.4-nano | — |
| API providers | 80 (best) | 36 | 25 |
| Released | Aug 5, 2025 | Mar 17, 2026 | Jul 21, 2025 |
| Knowledge cutoff | Jun 2024 | Aug 31, 2025 | — |
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-5.4 nano$4.50
Qwen3 235B-A22B Instruct 2507$2.70
Which should you choose?
Which is better: GPT OSS 120B, GPT-5.4 nano or Qwen3 235B-A22B Instruct 2507?
GPT-5.4 nano is the better all-round choice, scoring 68/100 against GPT OSS 120B (61) and Qwen3 235B-A22B Instruct 2507 (60). It leads on capability, inputs & features and context window. The score weighs capability 50%, price 25%, inputs & features 15%, context window 10%.
Which is cheaper, GPT OSS 120B, GPT-5.4 nano or Qwen3 235B-A22B Instruct 2507?
GPT OSS 120B is cheaper at $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); GPT-5.4 nano costs $0.20 input / $1.25 output per million tokens (official OpenAI API price). At a typical mix of three input tokens to one output token, that is $0.263 per million tokens for GPT OSS 120B versus $0.263 for Qwen3 235B-A22B Instruct 2507 (1× as much) and $0.463 for GPT-5.4 nano (1.8× as much).
Which scores higher on benchmarks?
GPT-5.4 nano scores higher on the Capabilities Index (ECI): GPT-5.4 nano 146.3 (#86 of 158), GPT OSS 120B 140.2 (#115 of 158) and Qwen3 235B-A22B Instruct 2507 138.1 (#120 of 158). Their confidence ranges do not overlap (144.7–148.0 vs 138.3–142.1), so the gap is a real one.
Which is better for coding?
There are no published SWE-bench Verified results for GPT OSS 120B, GPT-5.4 nano and Qwen3 235B-A22B Instruct 2507 yet, so there is no like-for-like coding score. On overall capability, GPT-5.4 nano 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?
GPT-5.4 nano has the largest context window at 400,000 tokens, against 262,144 for Qwen3 235B-A22B Instruct 2507 and 131,072 for GPT OSS 120B. Maximum output per response: GPT OSS 120B up to 131,072, GPT-5.4 nano up to 128,000, Qwen3 235B-A22B Instruct 2507 up to 16,384 tokens.
Which can read images, PDFs, audio or video?
GPT OSS 120B accepts text; GPT-5.4 nano accepts text and images; Qwen3 235B-A22B Instruct 2507 accepts text. GPT-5.4 nano handles the widest range of inputs.
Are any of these open source?
GPT OSS 120B and Qwen3 235B-A22B Instruct 2507 publishes its weights (Apache 2.0) and can be self-hosted; GPT-5.4 nano is proprietary.
Which is newer?
GPT-5.4 nano is the newest, released Mar 17, 2026. GPT OSS 120B came out Aug 5, 2025; Qwen3 235B-A22B Instruct 2507 came out Jul 21, 2025. Knowledge cutoff: GPT OSS 120B Jun 2024, GPT-5.4 nano Aug 31, 2025.
How do you decide the winner?
This rating uses capability (50%, independent benchmark results); price (25%, blended price per million tokens (3 input : 1 output), log scale); inputs & features (15%, image, PDF, audio and video input, tool calling, structured output and reasoning); context window (10%, 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.