DeepSeek V4 Flash (0731)
DeepSeek's fast, low-cost text tier, the 0731 release of V4 Flash: 1M context, thinking, and tool calling for high-volume, cost-sensitive workloads.
- Input
- $0.28 / 1M
- Output
- $0.56 / 1M
- Cached
- $0.07 / 1M
- Off-peak
- 50% off · 22:00–06:00 CET
- Context
- 1M
- Type
- Chat
- Maker
- DeepSeek
- Released
- 2026-08-01
Positioning
One of the lowest-cost text models in the sference catalog, with the cheapest output tokens, and the default pick for high-volume work where cost per token dominates: extraction, classification, summarisation, and background agent steps. For image input, DeepSeek V4.1 Flash or GLM 5.3 Flash cover the same ground with vision.
Compared to the portfolio
Artificial Analysis Intelligence Index v4.3 (0–100), measured across all providers by Artificial Analysis, not sference specifically. Blended price weights input and output 3:1. Hover a point for the reasoning-effort variant scored. Source
| Model | Context | Input | Output | Index | Vision |
|---|---|---|---|---|---|
| GLM 5.3 | 1M | $1.20 | $4.20 | 45 | — |
| DeepSeek V4.1 Flash (1017) | 1M | $0.35 | $1.32 | 45 | ✓ |
| GLM 5.3 Flash | 1M | $0.20 | $0.60 | 42 | ✓ |
| DeepSeek V4 Flash (0731) | 1M | $0.28 | $0.56 | 34 | — |
| Clef 27Bdecisions | 64K | $0.24 | — | — | ✓ |
Serverless rates per 1M tokens at the realtime (list) rate — priority bills 1.3×, flex 0.7×, and batch 0.5× of it, and requests that arrive off-peak bill at a discount (see off-peak pricing) — confirmed per model when your account is provisioned. Dedicated deployments are billed per GPU-hour on monthly commitments.
Use it
POST /v1/chat/completions- Realtime, priority, and flex (service_tier)
POST /v1/responses- Realtime, priority, flex, and the 24h batch window (background: true)
POST /v1/messages- Anthropic Messages shape, realtime only
curl https://api.sference.com/v1/chat/completions \
-H "Authorization: Bearer $SFERENCE_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "deepseek-ai/DeepSeek-V4-Flash-0731",
"messages": [{ "role": "user", "content": "Say hello in one sentence." }]
}'OpenAI-compatible; most clients switch with a base-URL change. Quickstart · Python SDK · Claude Code
Dedicated deployment
Run DeepSeek V4 Flash (0731) on GPU capacity reserved for you: a private endpoint, predictable latency, billed per GPU-hour on a monthly commitment. Tell us your expected volume and we come back with a sizing and a quote. How dedicated works
Minimum footprint 4 GPUs (80 GB) or 2 GPUs (141 GB) · License MIT
FAQ
What is the difference between `DeepSeek-V4-Flash` and `DeepSeek-V4-Flash-0731`?
deepseek-ai/DeepSeek-V4-Flash-0731 pins DeepSeek's 0731 release. deepseek-ai/DeepSeek-V4-Flash is an alias that currently serves it. Use the dated id if you want the version you validated to stay fixed; the alias can move to a newer DeepSeek release, which we announce by email first.
Should I use V4 Flash or V4.1 Flash?
V4 Flash is cheaper per input and output token and is text-only. V4.1 Flash, its successor, adds image input and a lower cached-input rate. Stay on V4 Flash for text-only, cost-sensitive workloads; move when you need vision. V4 Flash keeps running, and we will give notice with a date before it retires.
What precision is DeepSeek V4 Flash served at?
DeepSeek's native hybrid checkpoint, as published: MXFP4 for the MoE experts and FP8 (e4m3) for the rest of the weights. We add no extra quantization.
What can DeepSeek V4 Flash (0731) do on sference?
DeepSeek V4 Flash (0731) has a 1M-token context window and supports thinking and tool calling. It does not accept image input; for vision workloads, pick a model marked vision in the comparison table above. Turn on thinking with thinking: {"type": "enabled"}.
Which APIs can I call DeepSeek V4 Flash (0731) through?
POST /v1/chat/completions and POST /v1/responses (realtime, flex via service_tier: "flex", and the 24h async window), plus POST /v1/messages (Anthropic-shaped, realtime only). The OpenAI and Anthropic SDKs work by changing the base URL, and the same model id works on every endpoint. See the quickstart and the Anthropic Messages guide.
Is DeepSeek V4 Flash (0731) cheaper for background or batch work?
Yes. Flex (service_tier: "flex", enabled per account) bills 0.7× the realtime rate for requests that can wait up to 30 minutes, and the 24h batch window (batches and background responses) bills 0.5×. Latency-critical requests can go the other way: service_tier: "priority" jumps the queue at 1.3×. Off-peak pricing needs no flag at all: 50% off for requests that arrive 22:00–06:00 CET (21:00–05:00 UTC), so overnight agent runs and evals get the discount at realtime latency. Per 1M tokens: priority $0.364 in / $0.728 out, flex $0.196 in / $0.392 out, batch $0.14 in / $0.28 out, off-peak $0.14 in / $0.28 out. See pricing.
Where is data sent to DeepSeek V4 Flash (0731) processed, and is it used for training?
Every request runs on European GPUs. Inputs are never used for training, contractually and technically, and are deleted per your configured retention policy; compliance logs (model version, region, timestamps) are kept for your audit trail. Details in the DPA and sub-processor list.
What are the rate limits for DeepSeek V4 Flash (0731)?
There is no general per-account request limit, and the API does not send rate-limit headers. Image input has a per-team budget of 20 images per second; exceeding it returns 429. See Images and vision.
What happens when DeepSeek V4 Flash (0731) is updated or retired?
Pin an exact version by calling its versioned id: that version is the one that serves, and every response records it. Before a model retires we email you with a date, and the old id becomes an alias of its successor so existing integrations keep working. Rerun your evals against the successor before the date.
Generated from the live catalog (GET /v1/models), so pricing, context, and capabilities always match the API. Machine-readable version: https://sference.com/models/deepseek-ai/DeepSeek-V4-Flash-0731.md