GPT-5 Mini vs Qwen3 235B A22B Instruct 2507
GPT-5 Mini and Qwen3 235B A22B Instruct 2507 both landed mid-2025 and both target cost-conscious production use, but they solve different problems. GPT-5 Mini is OpenAI's compact reasoning model with multimodal input and mandatory reasoning; Qwen3 235B is a multilingual mixture-of-experts model (22B active params) built for straight text generation at a much lower price. The gap that matters most: input tokens cost 2.78x more on GPT-5 Mini.
Spec vs spec
| Spec | GPT-5 Mini | Qwen3 235B A22B Instruct 2507 |
|---|---|---|
| Context window | 400K | 262K |
| Max output | 128K | 16K |
| Input modalities | text, image, file | text |
| Output modalities | text | text |
| Knowledge cutoff | May 31, 2024 | Jun 30, 2025 |
| Released | Aug 7, 2025 | Jul 21, 2025 |
| Reasoning | always on | — |
Pricing
Per 1M tokens. Provider price plus the flat 3% fee — the sum is what you pay.
openai/gpt-5-mini
Input · 1M tokens
$0.250 + 3%$0.258
Output · 1M tokens
$2.00 + 3%$2.06
Cache read · 1M tokens
$0.025 + 3%$0.026
FEE — FLAT, EVERY MODEL3%
qwen/qwen3-235b-a22b-2507
Input · 1M tokens
$0.090 + 3%$0.093
Output · 1M tokens
$0.100 + 3%$0.103
FEE — FLAT, EVERY MODEL3%
One workload, priced on both
10M input + 2M output tokens at each model's price, flat 3% fee included.
openai/gpt-5-mini
$6.70
$6.50 provider + 3%
qwen/qwen3-235b-a22b-2507Cheaper
$1.13
$1.10 provider + 3%
Benchmarks
Design Arena categories where both models have results. Higher Elo and lower rank win.
| GPT-5 Mini | Qwen3 235B A22B Instruct 2507 | |||
|---|---|---|---|---|
| Category | Elo | Rank | Elo | Rank |
| 3D | 1114 | #73 | 1071 | #80 |
| Code | 1164 | #64 | 1088 | #81 |
| Data viz | 1167 | #62 | 1102 | #77 |
| Game dev | 1194 | #54 | 1018 | #94 |
| UI components | 1161 | #57 | 1022 | #86 |
| Websites | 1167 | #62 | 1101 | #83 |
Head-to-head preference voting. How we filter and rank
Pricing math
GPT-5 Mini runs $0.25/M input and $2.00/M output at the provider rate; on OpenKey that's $0.25 x 1.03 = $0.2575/M input and $2.00 x 1.03 = $2.06/M output. Qwen3 235B is $0.09/M input and $0.10/M output provider-side, or $0.0927/M and $0.103/M on OpenKey after the flat 3% fee.
Run the numbers on a 10M-input/2M-output workload: GPT-5 Mini totals $6.50, Qwen3 235B totals $1.10. That's the input_price_ratio of 2.78x showing up directly in your bill. If you're doing bulk text generation, classification, or RAG pipelines with heavy input volume, Qwen3's price advantage compounds fast. GPT-5 Mini also charges for cache reads at $0.025/M provider-side — Qwen3 has no listed cache pricing.
Context and modality
GPT-5 Mini offers 400,000 tokens of context against Qwen3's 262,144 — a 1.53x ratio in GPT-5 Mini's favor. But GPT-5 Mini caps completions at 128,000 tokens versus Qwen3's 16,384, so if you need long generated outputs (not just long inputs), GPT-5 Mini has 7.8x more headroom there.
Modality is the bigger differentiator: GPT-5 Mini accepts text, image, and file input and outputs text only. Qwen3 235B is text-to-text exclusively. If your workload involves parsing screenshots, scanned PDFs, or diagrams, Qwen3 is not an option — GPT-5 Mini is the only one of the two that can even attempt it.
Coding and UI generation benchmarks
On Design Arena's model benchmarks, GPT-5 Mini beats Qwen3 235B across every shared category. Code categories: GPT-5 Mini elo 1164 (rank 64) vs Qwen3 1088 (rank 81). Data visualization: 1167 (rank 62) vs 1102 (rank 77). Game dev: 1194 (rank 54) vs 1018 (rank 94) — the widest gap in the set. UI components: 1161 (rank 57) vs 1022 (rank 86). Website generation: 1167 (rank 62) vs 1101 (rank 83). 3D: 1114 (rank 73) vs 1071 (rank 80).
GPT-5 Mini also has an ASCII art category score (1173, rank 33) that Qwen3 wasn't scored on. For any front-end or generative-code task judged by these categories, GPT-5 Mini is ahead on every axis measured — the reasoning-effort controls likely help here.
Reasoning control and parameters
GPT-5 Mini has reasoning baked in as mandatory, with four effort levels (high, medium, low, minimal) and medium as default — useful when you want to trade latency for accuracy on a per-request basis. It supports `include_reasoning`, `reasoning`, and `structured_outputs` among its parameters.
Qwen3 235B has no reasoning mode listed but exposes a much larger classic sampling surface: `temperature`, `top_p`, `top_k`, `min_p`, `repetition_penalty`, `frequency_penalty`, `presence_penalty`, `logit_bias`, and `logprobs`. If your pipeline depends on fine-grained sampling control rather than reasoning depth, Qwen3's parameter set gives you more direct knobs.
When to pick each
Choose GPT-5 Mini for multimodal input, long single-document context (400K), controllable reasoning effort, or any of the coding/UI benchmark categories where it leads. Choose Qwen3 235B A22B for high-volume text generation, multilingual tasks, or any workload where the 2.78x input price gap and $1.10-vs-$6.50 cost delta on a 10M/2M workload actually changes your unit economics. Both models are available on OpenKey with a single API key, billed at provider list price plus a flat 3% fee.
Which model for which job
| Use case | Pick | Why |
|---|---|---|
| Bulk text classification/RAG at scale | Qwen3 235B A22B Instruct 2507 | $1.10 vs $6.50 on a 10M-in/2M-out workload — 5.9x cheaper |
| Parsing images, screenshots, or file uploads | GPT-5 Mini | Only GPT-5 Mini accepts image and file input; Qwen3 is text-only |
| Generative UI / website / game-dev code | GPT-5 Mini | Leads every Design Arena coding category, e.g. gamedev elo 1194 vs 1018 |
| Long single-document analysis (300K+ tokens) | GPT-5 Mini | 400K context vs Qwen3's 262,144 — 1.53x more room |
| Long-form generated output (novels, large reports) | GPT-5 Mini | 128,000 max completion tokens vs Qwen3's 16,384 |
| Multilingual text generation on a budget | Qwen3 235B A22B Instruct 2507 | Provider price of $0.09/M input, built as a multilingual MoE model |
Questions
- Which is cheaper, GPT-5 Mini or Qwen3 235B?
- Qwen3 235B, by a wide margin. On OpenKey, GPT-5 Mini costs $0.2575/M input and $2.06/M output; Qwen3 costs $0.0927/M input and $0.103/M output. A 10M-input/2M-output workload costs $6.50 on GPT-5 Mini versus $1.10 on Qwen3 — a 2.78x input price ratio.
- Which model has more context?
- GPT-5 Mini, with 400,000 tokens versus Qwen3's 262,144 — a 1.53x ratio. GPT-5 Mini also allows up to 128,000 completion tokens versus Qwen3's 16,384 cap, so it has more room on both ends.
- Can Qwen3 235B process images?
- No. Qwen3 235B A22B Instruct 2507 is text-to-text only. GPT-5 Mini accepts text, image, and file inputs, so it's the only choice between the two for multimodal workloads.
- Which model scores better on coding benchmarks?
- GPT-5 Mini, on every shared Design Arena category. Game dev is the widest gap: GPT-5 Mini elo 1194 (rank 54) versus Qwen3's 1018 (rank 94). GPT-5 Mini also leads code categories, data viz, UI components, website, and 3D.