Skip to content
OpenKey

GPT-5 Mini vs Qwen3 235B A22B Instruct 2507

OpenAIQwenboth via one key, provider price + 3%

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

SpecGPT-5 MiniQwen3 235B A22B Instruct 2507
Context window400K262K
Max output128K16K
Input modalitiestext, image, filetext
Output modalitiestexttext
Knowledge cutoffMay 31, 2024Jun 30, 2025
ReleasedAug 7, 2025Jul 21, 2025
Reasoningalways on

Pricing

Per 1M tokens. Provider price plus the flat 3% fee — the sum is what you pay.

openkey.ai

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%

openkey.ai

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 MiniQwen3 235B A22B Instruct 2507
CategoryEloRankEloRank
3D1114#731071#80
Code1164#641088#81
Data viz1167#621102#77
Game dev1194#541018#94
UI components1161#571022#86
Websites1167#621101#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 casePickWhy
Bulk text classification/RAG at scaleQwen3 235B A22B Instruct 2507$1.10 vs $6.50 on a 10M-in/2M-out workload — 5.9x cheaper
Parsing images, screenshots, or file uploadsGPT-5 MiniOnly GPT-5 Mini accepts image and file input; Qwen3 is text-only
Generative UI / website / game-dev codeGPT-5 MiniLeads every Design Arena coding category, e.g. gamedev elo 1194 vs 1018
Long single-document analysis (300K+ tokens)GPT-5 Mini400K context vs Qwen3's 262,144 — 1.53x more room
Long-form generated output (novels, large reports)GPT-5 Mini128,000 max completion tokens vs Qwen3's 16,384
Multilingual text generation on a budgetQwen3 235B A22B Instruct 2507Provider 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.

Go deeper