DeepSeek peak/off-peak pricing update
TL;DR Highlight
DeepSeek launched V4-Pro and set off-peak API rates 50% below peak
Who Should Read
A backend developer running batch jobs or agent workflows on the DeepSeek API who wants to cut costs
Core Mechanics
- V4-Pro is now generally available
- It is available in the app/web via "Expert Mode" and through the API.
- API model names are unchanged, so follow the setup guide in the docs.
- Reasoning effort has three levels
- Both V4-Pro and V4-Flash let you choose low, high, or max.
- Low is for simple tasks, high for daily agent workflows, and max for complex tasks.
- Agent performance reportedly improved a lot
- The post only says "strong production gains" and gives no numbers.
- Native OpenAI Responses API support
- It is optimized for Codex with one-click setup.
- Off-peak rates are 50% lower than peak
- New pricing takes effect at 16:00 UTC on Aug 16, 2026.
- The idea is to let you schedule flexible workloads.
- The post does not give actual prices or peak hours.
Community Reactions
How to Apply
- List your delay-tolerant jobs
- Spend 30 minutes today listing non-urgent calls such as nightly summaries, embedding preprocessing, and eval batches.
- Note the monthly call volume for each.
- Check peak hours on the pricing page
- Look up peak and off-peak hours and unit prices on the official pricing page.
- The post does not include them.
- Schedule jobs for off-peak in your queue
- Change your queue scheduler so delay-tolerant jobs run off-peak.
- Leave user-facing calls at peak pricing.
- Split reasoning effort by task
- Set simple calls such as classification or extraction to low.
- Use high for agent loops and max only for hard problems.
Try It
# Paste into Claude or ChatGPT (a quick audit prompt based on the pricing change)
Our service uses the DeepSeek API. Review the call list below and split the jobs into ones that can be deferred to off-peak and ones that cannot.
Assume only that off-peak rates are 50% lower than peak; you do not know the actual prices or hours.
Call list:
{{call list: name, daily calls, acceptable delay}}
For each job, answer in this format.
- Job name / off-peak OK? / reason / recommended reasoning effort (low, high, max)
Next step: add the monthly call count for each off-peak job and ask for the savings percentage versus peak pricing.Terminology
Related Papers
Mercury 3: Diffusion at scale
Scaling diffusion LLM 40x gives Mercury 3: 4–8x faster than speed-tier models, similar quality
Timestral: Mistral's time-series foundation models
A 209M time-series model ranks first on three benchmarks and runs on CPU.
DeepSeek V4.1 Flash 1017 (Max)
DeepSeek V4.1 Flash 1017 Max scores 45 on the Intelligence Index at 211 tok/s and $0.30 input.
If you don't run your own LLM proxy, how do you know whats going on?
Running your own LLM proxy shows which keys, models, and cache misses drive your usage.
DeepSeek: DeepSeek v4.1 Flash 1017
DeepSeek V4.1 Flash 1017: $0.05/M input, 1M context, built for agents
DeepSeek-v4.1-Flash-1017
DeepSeek V4.1 Flash minor update lifts coding and agent benchmarks while cutting run time and output length
Show HN: Llmspeed – feel how fast a local LLM runs before you buy the hardware