DeepSeek V4 Flash API
The official DeepSeek-V4-Flash release: a 1,048,576-token context window at a workhorse price, and genuinely fast — about 1 second to first token and ~100 tokens/second in our measurements. OpenAI-compatible, Cloudflare-hosted, 40% off list for the launch period.
Specs
- Context window
- 1,048,576 tokens (input + max_tokens)
- Function calling
- Yes — calls tools with correct schemas in our contract tests
- Reasoning
- Yes — reasoning_effort accepted; default low via the gateway
- Upstream ID (public)
- @cf/deepseek-ai/deepseek-v4-flash-0731
- Privacy
- Cloudflare-hosted; metadata-only logging on our side
Pricing (40% off list)
- Input
$0.44$0.264- Output
$1.32$0.792
@cf/deepseek-ai/deepseek-v4-flash-0731
Struck price = list. Full live pricing, including cached input, on the models page.
Measured speed
Our measurements, 2026-08-16 — the reason we call Flash fast is that we checked.
~1.0 s
time to first token (p50, 5 streams)
~102 tok/s
streaming throughput (p50)
4.6 s
first token at 100K input tokens
13.2 s
first token at 300K input tokens (~45 ms per 1K tokens)
Quickstart
OpenAI SDK unchanged — set base_url, use model id deepseek-v4-flash.
from openai import OpenAI
client = OpenAI(
base_url="https://api.clfaigateway.dev/v1",
api_key="sk-gw-...", # app.clfaigateway.dev
)
resp = client.chat.completions.create(
model="deepseek-v4-flash",
messages=[{"role": "user", "content": "Summarize this repository..."}],
stream=True,
)
for chunk in resp:
print(chunk.choices[0].delta.content or "", end="")deepseek-v4-flash — common questions
Long-context agent loops, repository and document Q&A, extraction and classification at volume — anywhere you want the 1M window and real speed without flagship pricing. It calls tools correctly and streams its reasoning.
Same 1M window and the same tokenizer; Flash is about a third of Pro’s price, much faster to first token in our tests, and the right default. Escalate the hardest reasoning to Pro when quality on a specific task demands it.
deepseek-v4-pro →Very large requests (600K+ tokens) can hit retryable capacity errors at peak times, and per-minute rate limits are tighter on frontier-class models than on small ones. The gateway returns honest 429s with retry-after; official SDKs back off automatically.
Rate-limit docs →