DS
Reasoning / generalist / open-weight / self-host
DeepSeek
-
V4‑Pro‑0813 from 08/13/2026 peak $1.32/$3.96
1M context, output up to 384k, MIT license. Off‑peak, since 08/16/2026, is $0.66/$1.98.
-
V4.1 Flash off-peak about $0.15/$0.60
The current Flash, with vision. The older deepseek‑v4‑flash slug already serves V4.1.
Peak hours are 01:00-04:00 and 06:00-10:00 UTC. V3 and R1 are being retired.
- Verified2026-09-20
- Models covered3
- Sources3
- Deployment channels2
Purchase decision (when to choose / when to avoid)
Choose if...
- Priority is reasoning/code at low token cost (high volume).
- You want MIT / no MAU clauses - easier legal compliance for self-host.
- You're building batch processing and optimizing cache/costs.
Avoid if...
- You need the largest enterprise ecosystem and ready EU integrations.
- You have use cases requiring top Polish quality - consider Bielik or top models with PL tests.
Cost in practice (scenarios)
High volume
Usually very competitive token cost; caching pays off.
- batch
- automations
Self-host
MIT simplifies legal; cost is GPU + maintenance.
- MLOps
These are estimates/scenarios (not an invoice). Actual cost depends on context length, number of users, limits and retention policies.
Deployment / data / enterprise
Deployment channels
- DeepSeek API (OpenAI-compatible)
- Self-host (MIT) - depending on variant and resources
Data policy
- Training on data
- Depends on mode (API vs self-host).
- Retention
- API: depends on terms; self-host: on your side.
- Data residency
- Depends on region/service.
For companies: key are retention policies and regions - check sources.
Enterprise readiness
- Admin
- API + billing; enterprise depends on offering.
- SSO/SCIM
- Depends on offering.
- Audit
- Depends on offering.
- DPA
- Depends on agreement.
- Certifications
- Depends on agreement.
Great for volume reasoning/code, but enterprise integrations can be modest.