DeepSeek V3.2 vs Qwen3 235B A22B Instruct 2507
Both models are text-only, similarly priced at the low end of the market, and released within months of each other — DeepSeek V3.2 in December 2025, Qwen3 235B A22B Instruct 2507 in July 2025. The difference that matters is what you're optimizing for: DeepSeek V3.2 leads on every shared Design Arena coding/design category, while Qwen3 235B A22B costs less per token and doubles the usable context window. Both run on OpenKey under one API key, billed at provider list price plus a flat 3% fee.
Spec vs spec
| Spec | DeepSeek V3.2 | Qwen3 235B A22B Instruct 2507 |
|---|---|---|
| Context window | 131K | 262K |
| Max output | 64K | 16K |
| Input modalities | text | text |
| Output modalities | text | text |
| Knowledge cutoff | — | Jun 30, 2025 |
| Released | Dec 1, 2025 | Jul 21, 2025 |
| Reasoning | optional | — |
Pricing
Per 1M tokens. Provider price plus the flat 3% fee — the sum is what you pay.
deepseek/deepseek-v3.2
Input · 1M tokens
$0.229 + 3%$0.236
Output · 1M tokens
$0.343 + 3%$0.353
Cache read · 1M tokens
$0.023 + 3%$0.024
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.
deepseek/deepseek-v3.2
$3.06
$2.97 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.
| DeepSeek V3.2 | Qwen3 235B A22B Instruct 2507 | |||
|---|---|---|---|---|
| Category | Elo | Rank | Elo | Rank |
| 3D | 1210 | #41 | 1071 | #80 |
| Code | 1213 | #48 | 1088 | #81 |
| Data viz | 1203 | #48 | 1102 | #77 |
| Game dev | 1197 | #50 | 1018 | #94 |
| UI components | 1203 | #47 | 1022 | #86 |
| Websites | 1217 | #46 | 1101 | #83 |
Head-to-head preference voting. How we filter and rank
Pricing math on a real workload
DeepSeek V3.2 prices at $0.2288/M input and $0.3432/M output from the provider; on OpenKey that's $0.235664/M input and $0.353496/M output after the 3% fee ($0.2288 × 1.03 and $0.3432 × 1.03). Qwen3 235B A22B Instruct 2507 is $0.09/M input and $0.10/M output from the provider, or $0.0927/M and $0.103/M on OpenKey.
On a 10M-input / 2M-output token workload, DeepSeek V3.2 costs $2.97 total versus $1.10 for Qwen3 235B — a 2.7x gap driven mostly by input tokens (DeepSeek's input price is 2.54x Qwen3's). DeepSeek also offers a provider cache-read rate of $0.02288/M, which Qwen3 doesn't list, so repeated-context workloads narrow the gap somewhat for DeepSeek.
Design Arena results
DeepSeek V3.2 outranks Qwen3 235B in every category both models were scored in: 3d (elo 1210, rank 41 vs elo 1071, rank 80), codecategories (1213/rank 48 vs 1088/rank 81), dataviz (1203/rank 48 vs 1102/rank 77), gamedev (1197/rank 50 vs 1018/rank 94), uicomponent (1203/rank 47 vs 1022/rank 86), and website (1217/rank 46 vs 1101/rank 83). The gap is largest in gamedev and uicomponent, where Qwen3's win rate drops into the mid-30s to high-30s versus DeepSeek's high-40s to low-50s. DeepSeek V3.2 also has scores for asciiart (1129, rank 42) and svg (1089, rank 54) where Qwen3 has no listed result. If your workload is UI generation, game logic, or front-end scaffolding, this isn't a close call.
Context and output limits
Qwen3 235B A22B Instruct 2507 supports a 262,144-token context window, exactly double DeepSeek V3.2's 131,072 tokens. That matters for long-document summarization, large codebase review, or multi-file agent context. DeepSeek V3.2 counters with a larger max completion: 64,000 tokens versus Qwen3's 16,384. So DeepSeek can generate longer single responses, while Qwen3 can ingest more before it needs to truncate or chunk. If your task is 'read a huge repo, answer briefly,' Qwen3's context wins. If it's 'generate a long file or multi-step output,' DeepSeek's completion ceiling wins.
Tool use and structured output
Both models support `tools`, `tool_choice`, `response_format`, and `structured_outputs` — so both work fine in agent loops or JSON-mode pipelines. DeepSeek V3.2 additionally supports `reasoning` and `include_reasoning` parameters (reasoning is optional, not on by default), giving you a lever to trade latency for deeper multi-step thinking on harder tasks. Qwen3 235B A22B Instruct 2507 has no reasoning parameter exposed — it's a straight instruction-following model. If your agent needs occasional chain-of-thought on hard sub-tasks without switching models, DeepSeek gives you that toggle.
Which model for which job
| Use case | Pick | Why |
|---|---|---|
| UI component / front-end generation | DeepSeek V3.2 | elo 1203 vs 1022 on uicomponent, rank 47 vs 86 |
| Long-document ingestion (large codebase, big PDFs) | Qwen3 235B A22B Instruct 2507 | 262,144-token context, double DeepSeek's 131,072 |
| High-volume batch processing on a budget | Qwen3 235B A22B Instruct 2507 | $1.10 vs $2.97 on the same 10M-in/2M-out workload |
| Game logic / gamedev prototyping | DeepSeek V3.2 | elo 1197 vs 1018, rank 50 vs 94 |
| Long single-response generation (full file dumps) | DeepSeek V3.2 | 64,000-token max completion vs Qwen3's 16,384 |
| Agent tasks needing occasional deep reasoning | DeepSeek V3.2 | optional `reasoning` parameter available; Qwen3 has none |
Questions
- Which model is cheaper for a typical 10M input / 2M output job?
- Qwen3 235B A22B Instruct 2507, at $1.10 total on OpenKey versus $2.97 for DeepSeek V3.2 — about 2.7x cheaper, mainly because Qwen3's input price ($0.0927/M) is less than half DeepSeek's ($0.235664/M).
- Which model has a bigger context window?
- Qwen3 235B A22B Instruct 2507, with 262,144 tokens versus DeepSeek V3.2's 131,072 — exactly double, per the context ratio in the data.
- Does either model score better on coding-related benchmarks?
- Yes. DeepSeek V3.2 beats Qwen3 235B in every shared Design Arena category, including codecategories (elo 1213, rank 48 vs elo 1088, rank 81) and uicomponent (1203, rank 47 vs 1022, rank 86).
- Can I call both models through the same API key?
- Yes — both are available on OpenKey, which routes to 329 models across 52 labs through one key, charging provider list price plus a flat 3% fee (so DeepSeek V3.2's $0.2288/M input becomes $0.235664/M, and Qwen3's $0.09/M becomes $0.0927/M).