ERNIE 4.5 21B A3B vs GPT OSS 20B vs Qwen3 30B-A3B Instruct 2507
GPT OSS 20B comes out ahead, 61 to 57 and 57 on our weighted score, and it is the cheaper option too.
BaiduERNIE 4.5 21B A3B
57/100- ECI—
- Price$0.07 / $0.28
- Context131K
- Our pick
OpenAIGPT OSS 20B
61/100- ECI137.8
- Price$0.07 / $0.25
- Context131K
Alibaba (Qwen)Qwen3 30B-A3B Instruct 2507
57/100- ECI137.3
- Price$0.10 / $0.30
- Context262K
GPT OSS 20B is our pick
GPT OSS 20B is the better all-round choice, scoring 61/100 against Qwen3 30B-A3B Instruct 2507 (57) and ERNIE 4.5 21B A3B (57). It leads on inputs & features. Qwen3 30B-A3B Instruct 2507 wins on context window. The score weighs price 50%, inputs & features 30%, context window 20%. None of the shared benchmarks cover every model here yet, so capability is left out of this verdict.
- CapabilityNot enough dataNo independent benchmark covers every model here yet
- Lowest priceGPT OSS 20BGPT OSS 20B $0.115 · ERNIE 4.5 21B A3B $0.122 · Qwen3 30B-A3B Instruct 2507 $0.15 per 1M tokens (3:1 blend)
- Longest contextQwen3 30B-A3B Instruct 2507Qwen3 30B-A3B Instruct 2507 262,144 · ERNIE 4.5 21B A3B 131,072 · GPT OSS 20B 131,072 tokens
- Widest inputsSame inputsERNIE 4.5 21B A3B: Text · GPT OSS 20B: Text · Qwen3 30B-A3B Instruct 2507: Text
- Self-hostingAll open weightsEvery model here can be downloaded and run on your own hardware
| Measure | Weight | ERNIE 4.5 21B A3B | GPT OSS 20B | Qwen3 30B-A3B Instruct 2507 |
|---|---|---|---|---|
| Price | 50% | 89 | 90 | 85 |
| Inputs & features | 30% | 25 | 35 | 25 |
| Context window | 20% | 25 | 25 | 37 |
| Overall | 100% | 57/100 | 61/100 | 57/100 |
Left out because at least one model lacks the data: capability. The remaining weights were rescaled.
Every spec in one table
Highlighted cells lead their row. Dashes mean the data is not published.
| Specification | ERNIE 4.5 21B A3BBaidu | GPT OSS 20BOpenAI | Qwen3 30B-A3B Instruct 2507Alibaba (Qwen) |
|---|---|---|---|
| Capability | |||
| Capabilities Index (ECI) | — | 137.8 (best) | 137.3 |
| ECI rank | — | #122 of 165 (best) | #124 of 165 |
| GPQA DiamondGraduate-level science questions | — | 60.8% (best) | 55.6% |
| OTIS Mock AIME 2024–2025Competition mathematics | — | 65.3% (best) | 62.2% |
| Price per million tokens | |||
| Input | $0.07 (best) | $0.07 (best) | $0.10 |
| Output | $0.28 | $0.25 (best) | $0.30 |
| Cached input | — | — | — |
| Blended (3:1) | $0.122 | $0.115 (best) | $0.15 |
| Long-context rate | Same rate | Same rate | Same rate |
| Price source | Median of 1 providers | Median of 37 providers | Median of 12 providers |
| Limits | |||
| Context window | 131,072 tokens | 131,072 tokens | 262,144 tokens (best) |
| Max output | 8,000 tokens | 32,768 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 | No | Yes | No |
| Tool calling | Yes | Yes | Yes |
| Structured output | No | No | No |
| Availability | |||
| Weights | OpenApache-2.0 | Open | OpenApache 2.0 |
| API model ID | — | — | — |
| API providers | 1 | 41 (best) | 14 |
| Released | Jun 30, 2025 | Aug 5, 2025 | Jul 28, 2025 |
| Knowledge cutoff | — | 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.
- ERNIE 4.5 21B A3B$1.26
- GPT OSS 20B$1.20
- Qwen3 30B-A3B Instruct 2507$1.60
Which should you choose?
Which is better: ERNIE 4.5 21B A3B, GPT OSS 20B or Qwen3 30B-A3B Instruct 2507?
GPT OSS 20B is the better all-round choice, scoring 61/100 against Qwen3 30B-A3B Instruct 2507 (57) and ERNIE 4.5 21B A3B (57). It leads on inputs & features. Qwen3 30B-A3B Instruct 2507 wins on context window. The score weighs price 50%, inputs & features 30%, context window 20%. None of the shared benchmarks cover every model here yet, so capability is left out of this verdict.
Which is cheaper, ERNIE 4.5 21B A3B, GPT OSS 20B or Qwen3 30B-A3B Instruct 2507?
GPT OSS 20B is cheaper at $0.07 input / $0.25 output per million tokens (median across 37 API providers). ERNIE 4.5 21B A3B costs $0.07 input / $0.28 output per million tokens (median across 1 API provider); Qwen3 30B-A3B Instruct 2507 costs $0.10 input / $0.30 output per million tokens (median across 12 API providers). At a typical mix of three input tokens to one output token, that is $0.115 per million tokens for GPT OSS 20B versus $0.122 for ERNIE 4.5 21B A3B (1.1× as much) and $0.15 for Qwen3 30B-A3B Instruct 2507 (1.3× as much).
Which scores higher on benchmarks?
There is no independent benchmark that covers all three models yet. ERNIE 4.5 21B A3B has not been scored yet, GPT OSS 20B has an ECI of 137.8 and Qwen3 30B-A3B Instruct 2507 has an ECI of 137.3.
Which is better for coding?
There are no published SWE-bench Verified results for ERNIE 4.5 21B A3B, GPT OSS 20B and Qwen3 30B-A3B Instruct 2507 yet, so there is no like-for-like coding score. Test both on a sample of your own tasks. All three support tool calling for agent workflows.
Which has the bigger context window?
Qwen3 30B-A3B Instruct 2507 has the largest context window at 262,144 tokens, against 131,072 for ERNIE 4.5 21B A3B and 131,072 for GPT OSS 20B. Maximum output per response: ERNIE 4.5 21B A3B up to 8,000, GPT OSS 20B up to 32,768, Qwen3 30B-A3B Instruct 2507 up to 16,384 tokens.
Which can read images, PDFs, audio or video?
ERNIE 4.5 21B A3B accepts text; GPT OSS 20B accepts text; Qwen3 30B-A3B 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 and Apache 2.0), so you can self-host them.
Which is newer?
GPT OSS 20B is the newest, released Aug 5, 2025. Qwen3 30B-A3B Instruct 2507 came out Jul 28, 2025; ERNIE 4.5 21B A3B came out Jun 30, 2025. Knowledge cutoff: GPT OSS 20B Jun 2024.
How do you decide the winner?
Each model gets a 0–100 score on 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 2, 2026.
