DeepSeek V4 Flash vs Qwen3-32B
Side-by-side comparison of API pricing, specs, benchmarks, and capabilities
| Specification |
by DeepInfra
3 providers:
DeepSeek $0.440/$1.32
Fireworks $0.140/$0.280
DeepInfra $0.090/$0.180
|
by DeepInfra
3 providers:
Alibaba $0.160/$0.640
Groq $0.290/$0.590
DeepInfra $0.080/$0.280
|
|---|---|---|
| Overview | ||
| Status | Current budget | Current budget Open weights |
| Released | Apr 24, 2026 | Jan 1, 2026 |
| Pricing per million tokens | ||
| Input | $0.090/Mtok | $0.080/Mtok |
| Output | $0.180/Mtok | $0.280/Mtok |
| Blended avg | $0.113/Mtok | $0.130/Mtok |
| Cached input | $0.018/Mtok | — |
| Price basis | DeepInfra lists two entries for this model and prices them differently: the floating alias DeepSeek-V4-Flash at $0.09 input / $0.018 cached input / $0.18 output per 1M tokens, and the pinned snapshot DeepSeek-V4-Flash-0731 at $0.08 / $0.016 / $0.18. The tracked figure is the floating alias, because that is what a caller who names the model without a date gets. Pinning to the 0731 snapshot saves 11% on input at the cost of being left behind when DeepInfra moves the alias. | — |
| Specifications | ||
| Context window | 1M tokens | 128K tokens |
| Parameters | Proprietary | 32B |
| Speed (TPS) | — | — |
| Modalities | ||
| Input |
text
|
text
|
| Benchmarks sources: DeepSeek model card / LMSYS Chatbot Arena (UC Berkeley) | ||
| Independent composite scores measured for both models — each on its own scale | ||
| AA Intelligence Index | 51.8 | 11.4 |
| AA Agentic Index | 48.4 | 1.8 |
| AA-Omniscience Index (−100–100) | -14.3 | -50.4 |
| GDPval-AA v2 | 1558.4 ELO | — |
| Per-benchmark results published for DeepSeek V4 Flash only — no independent per-benchmark scores exist for Qwen3-32B yet | ||
| Avg benchmark score | 77.5 | — |
| Perf / dollar | 688.9 | — |
| GPQA Diamond | 88.1 | — |
| SWE-Bench Verified | 79 vendor | — |
| LiveCodeBench | 91.6 | — |
| MCP Atlas | 69 | — |
| BrowseComp | 85.9 | — |
| Humanity's Last Exam | 51.6 | — |
| ARC-AGI 2 | 18 vendor | — |
| AIME 2025 | 58 vendor | — |
| MMMLU | 75 vendor | — |
| BFCL | 62 vendor | — |
| HumanEval | 79 | — |
| MATH 500 | 68 vendor | — |
| Providers | ||
| Available from |
DeepSeek — $0.440/$1.32/Mtok
Fireworks — $0.140/$0.280/Mtok
DeepInfra — $0.090/$0.180/Mtok
|
Alibaba — $0.160/$0.640/Mtok
Groq — $0.290/$0.590/Mtok
DeepInfra — $0.080/$0.280/Mtok
|
Cost at scale
1M tokens · 50/50 input/output| Volume | DeepSeek V4 Flash | Qwen3-32B | Savings |
|---|---|---|---|
| 1M tokens | $0.11 | $0.13 | $0.02 (15.4%) |
| 10M tokens | $1.13 | $1.3 | $0.18 (13.8%) |
| 100M tokens | $11.25 | $13 | $1.75 (13.5%) |
| 1000M tokens | $112.5 | $130 | $17.5 (13.5%) |
When to pick which
distilled from the pricing and spec data above- cost dominates: $0.113/Mtok blended vs $0.130 — 13.5% less on the same 50/50 token mix
- your workload re-reads context (agents, RAG, long chats): cached input costs $0.018/Mtok — 80% off its list input price, a discount Qwen3-32B doesn't offer
- you need the longer context: 1M tokens vs 128K (7.8×)
- you want open weights — self-host it, fine-tune it, or exit the API entirely; DeepSeek V4 Flash is closed
Summary
DeepSeek V4 Flash by DeepInfra costs $0.090/Mtok input and $0.180/Mtok output, with a 1M-token context window. It supports text input and is available from 3 providers.
Qwen3-32B by DeepInfra costs $0.080/Mtok input and $0.280/Mtok output, with a 128K-token context window. It supports text input and is available from 3 providers.
On a blended cost basis, DeepSeek V4 Flash is 13.5% cheaper than Qwen3-32B.It also has a larger context window.
The two aren't directly comparable on average benchmark score: DeepSeek V4 Flash has published per-benchmark results, while Qwen3-32B does not yet — it is measured today on independent composites (see the table above).
Note: Pricing is per million tokens. Actual costs vary with usage patterns, prompt caching, and batch discounts. Always verify against official provider pricing pages.