DeepSeek V4 Flash on QuickSilver Pro
DeepSeek V4 Flash 0731 is the official public-beta agent model: 1M-token context, stronger coding and tool use, and a native Responses API adapted for Codex. QuickSilver Pro serves the official DeepSeek build at FP8 precision or better for $0.06 / $0.18 per million tokens.
At a glance
Codex, agentic coding, tool use, and long-context automation.
Pricing comparison ($/1M tokens)
| Provider | Input | Output | vs QSP |
|---|---|---|---|
| QuickSilver Pro | $0.06 | $0.18 | — |
| OpenRouter (deepseek/deepseek-v4-flash-0731) | $0.06 | $0.18 | same |
| OpenAI (GPT-4o-mini) | $0.15 | $0.60 | 70% lower |
When to use
Default to V4 Flash for Codex and agentic coding, multi-step tool use, code generation, structured output, and long-context automation. The official 0731 build supports both OpenAI Chat Completions and Responses API clients while keeping the stable QSP model ID.
When to use something else
For genuinely hard reasoning (competitive programming, multi-step proofs, complex math), escalate to V4 Pro. For Opus-class agentic / long-horizon planning, Kimi K2.6 is the better fit. For audio or GPT-4-class creative writing, stay on OpenAI. V4 Flash itself is text-only — if you need image input, 38 of our 83 models accept it.
Quickstart (curl)
curl https://api.quicksilverpro.io/v1/chat/completions \
-H "Authorization: Bearer $QSP_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "deepseek-v4-flash",
"messages": [{"role": "user", "content": "Hello!"}]
}'OpenAI-compatible. One-line migration via base_url.
FAQ
Yes — V4 Flash thinks by default. For direct non-thinking chat through Chat Completions, pass `reasoning: { enabled: false }`; Responses API clients can set `reasoning.effort` to control depth.
$0.06 input / $0.18 output and $0.015 cache read per million tokens, matching the route-selected OpenRouter endpoint (deepinfra fp8) with no markup or discount. The route uses its pinned provider order or the cheapest eligible endpoint under sort: price and its quantization filters; the /models headline can differ. More expensive fallback hosts do not change your price. Use base_url=https://api.quicksilverpro.io/v1 and model="deepseek-v4-flash".
Yes. Use base_url=https://api.quicksilverpro.io/v1 with model="deepseek-v4-flash". QSP exposes both `/v1/chat/completions` and `/v1/responses`; Codex should use `wire_api = "responses"`.