Version
- V4 Flash
- DeepSeek-V4-Flash-0731
- V4 Pro
- DeepSeek-V4-Pro
DeepSeek V4 Flash guide
DeepSeek V4 Flash: API, pricing and setup
DeepSeek V4 Flash is the official lower-cost V4 API path for fast agent and coding workloads. As of August 3, 2026, it is also the only V4 model supported by DeepSeek's Responses API. The model is in public beta, so compare Flash with Pro and verify the interface your tool requires before migrating.
Answer first
Last checked: August 3, 2026
Status
Official API public beta
Model ID
deepseek-v4-flash
Current version
DeepSeek-V4-Flash-0731
Best default
Fast agent and coding iteration
What it is
A distinct V4 API route, not a generic alias
DeepSeek announced V4 Flash as a public beta API update on July 31, 2026. The stable model ID deepseek-v4-flash routes to the current DeepSeek-V4-Flash-0731 version.
The 0731 version statement applies to the API model. It does not prove that DeepSeek's app or web experience is running the same version.
DECLAREDOfficially declared API availability; independent production behavior is not claimed here.
Flash vs Pro
Choose by interface, workload and cost
Both are explicit V4 API models. Flash is the lower-cost, higher-concurrency option; Pro remains available for workloads that justify its different profile.
Positioning
- V4 Flash
- Lower-cost path for fast agent and coding workloads
- V4 Pro
- Higher-cost path for more demanding work
Context window
- V4 Flash
- 1M tokens
- V4 Pro
- 1M tokens
Maximum output
- V4 Flash
- 384K tokens
- V4 Pro
- 384K tokens
JSON output
- V4 Flash
- Yes
- V4 Pro
- Yes
Tool calling
- V4 Flash
- Yes
- V4 Pro
- Yes
Responses API
- V4 Flash
- Yes
- V4 Pro
- No as of August 3, 2026
Anthropic-compatible API
- V4 Flash
- Yes
- V4 Pro
- Yes
Cache-hit input
- V4 Flash
- $0.0028 / 1M tokens
- V4 Pro
- $0.003625 / 1M tokens
Cache-miss input
- V4 Flash
- $0.14 / 1M tokens
- V4 Pro
- $0.435 / 1M tokens
Output
- V4 Flash
- $0.28 / 1M tokens
- V4 Pro
- $0.87 / 1M tokens
Concurrency
- V4 Flash
- 2,500
- V4 Pro
- 500
Pricing status
These are current official standard API prices and may change. DeepSeek has previewed future peak and off-peak pricing: all billed items would be 2× during Beijing time 09:00–12:00 and 14:00–18:00. The effective date is pending an official notice; this schedule is not live as of August 3, 2026.
Official API setup
Start with the route your tool actually uses
DeepSeek offers OpenAI-compatible and Anthropic-compatible interfaces. Responses API support has a narrower model boundary.
Minimal OpenAI-compatible values
base_url=https://api.deepseek.com
model=deepseek-v4-flash
api_key=<your DeepSeek API Key>Chat Completions
Flash and Pro are listed for the OpenAI-compatible Chat Completions route.
Anthropic API
Use https://api.deepseek.com/anthropic. Both Flash and Pro are listed.
Responses API
Only V4 Flash is supported as of the review date. Pro support was previewed for early August, not confirmed live.
Model ID migration
Retired aliases need an explicit replacement
deepseek-chat and deepseek-reasoner were fully retired after July 24, 2026 at 15:59 UTC. Do not use either alias in a new production configuration.
Retired alias
deepseek-chat
Current path
deepseek-v4-flash or deepseek-v4-pro
Choose the explicit V4 model that matches the workload and interface you have verified.
Retired alias
deepseek-reasoner
Current path
Current model with Thinking Mode
Follow the current Thinking Mode documentation instead of treating the retired alias as a production configuration.
Coding and agent tools
Match the tool to its official interface
Official Anthropic-compatible route
Claude Code
DeepSeek maps claude-opus* requests to V4 Pro and haiku*/sonnet* requests to V4 Flash. The official example uses Pro for the main model and Flash for Haiku or subagent work.
Official Claude Code guideOfficial Responses API route
Codex
As of the review date, the Responses API supports V4 Flash only. Use the official Codex instructions and do not substitute a Chat Completions configuration.
Official Codex guideOfficial integration guide
OpenCode
DeepSeek recommends OpenCode 1.14.24 or later. Its guide connects the DeepSeek provider and selects Pro; verify the live model list before choosing Flash.
Official OpenCode guideTreat the official mappings as setup profiles, not universal quality rankings. Test Flash and Pro against your repository, tool-use pattern, latency target and budget before standardizing a team configuration.
Official API vs provider
Keep the delivery route visible
Same model name does not prove the same route, version, privacy terms or billing behavior.
A provider can be useful for payment, regional access or consolidated infrastructure, but its claims are not Modroam verification or a recommendation.
Official DeepSeek API
DeepSeek issues the key, publishes the endpoint and model interface, and controls the direct account terms.
Third-party provider
The provider controls its delivery route, billing and contract. Exact version and policy details require separate evidence.
- Who issued the API key, and which endpoint receives the request?
- Which exact model and version are delivered through which API interface?
- How are input, output, cache hits, concurrency and failed requests billed?
- What do the contract and policy say about retention, logs, training, region and zero-data-retention claims?
- Which payment, support and refund terms apply to this account?
Before you migrate
Five production checks
- 1Replace retired aliases with an explicit V4 Flash or V4 Pro model ID.
- 2Confirm whether the integration uses Chat Completions, Anthropic API or Responses API.
- 3Re-test tool calls, JSON output, long context and response handling.
- 4Recalculate spend with cache-hit, cache-miss, output and concurrency behavior.
- 5Review the live update, pricing and account terms before production traffic.
Access request
Need DeepSeek access?
Tell us whether you need platform access, an API route, a coding plan or setup support. We’ll check availability, scope and a test path.