AI agents · OpenClaw · self-hosting · automation

Quick Answer

DeepSeek V4 GA July 24: Surge Pricing & Migration Guide

Published:

DeepSeek V4 GA July 24: Surge Pricing & Migration Guide

DeepSeek V4 transitions from preview to General Availability on July 24, 2026 — and the same day, the legacy deepseek-chat and deepseek-reasoner aliases are retired at 15:59 UTC. GA also brings the API industry’s first structural peak/off-peak surge pricing for a frontier-tier model. If you call DeepSeek in production, here’s exactly what to change.

Last verified: July 23, 2026

The Two Changes That Matter

1. Legacy aliases are being retired

Old aliasRouted to (since Apr 24, 2026)Action required
deepseek-chatdeepseek-v4-flash (non-thinking)Switch to deepseek-v4-flash
deepseek-reasonerdeepseek-v4-flash (thinking)Switch to deepseek-v4-flash or deepseek-v4-pro

After 15:59 UTC, July 24, 2026, calls using the old names return HTTP errors. Many teams have unknowingly been on V4-Flash since April via transparent routing — GA just makes the explicit names mandatory.

2. Peak/off-peak surge pricing arrives

DeepSeek V4 GA is the first frontier-tier API to charge by time of day (UTC+8):

  • Peak hours: 09:00-12:00 and 14:00-18:00, seven days a week
  • Off-peak: overnight and weekends (UTC+8)
ModelInput (cache miss)Output off-peakOutput peak
DeepSeek V4 Pro~$0.435/MTok~$0.87/MTok~$1.74/MTok
DeepSeek V4 Flash~$0.028/MTok (miss)~$0.28/MTok~$0.56/MTok

Automatic disk-based prefix caching can cut input costs by up to ~98% for repeated prompts (V4 Flash cache-hit input as low as ~$0.0028/MTok).

Migration Checklist

  1. Find every call using deepseek-chat or deepseek-reasoner.
  2. Map them: general work → deepseek-v4-flash; hardest reasoning → deepseek-v4-pro.
  3. Restore thinking if neededdeepseek-reasoner was Flash-in-thinking-mode, not Pro. If you wanted Pro-grade reasoning, set deepseek-v4-pro and add extra_body={"thinking": {"type": "enabled"}}.
  4. Deploy before 15:59 UTC, July 24, 2026.
  5. Optimize cost: batch heavy jobs into off-peak UTC+8 windows and lean on prefix caching for repeated prompts.

Flash vs Pro — Quick Guidance

Use casePick
High-volume chat, extraction, classificationV4 Flash
Agentic pipelines (speed/cost sensitive)V4 Flash
Complex reasoning, instruction-heavy tasksV4 Pro
Full 1M-token context / long-horizon workV4 Pro
  • V4 Flash: 284B total / ~13B active MoE
  • V4 Pro: 1.6T total / ~49B active MoE, up to 1M context

Why DeepSeek Did This

Surge pricing is a load-balancing lever: DeepSeek wants to smooth cluster demand across time zones instead of buying more capacity for peak spikes. For developers, it’s an opportunity — schedule non-interactive work off-peak and your effective cost drops by roughly half on output tokens.

Sources