Is DeepSeek Still Good Value After the August 2026 Price Increase?
DeepSeek V4 Flash and Pro now use peak and off-peak API pricing. Compare the new rates with GPT-5.6 Luna, Mistral Small 4, Gemini 3.7 Flash, and GLM-5.2.
DeepSeek is still price-competitive, but it is no longer the automatic cheapest API for every text workload. The new V4 prices took effect at 2026-08-16 16:00 UTC. DeepSeek now charges a peak rate during 01:00-04:00 and 06:00-10:00 UTC, with a 50% off-peak discount at all other times.
The practical answer is conditional: V4 Flash remains attractive off-peak and for cache-heavy 1M-context agents, but Mistral Small 4 can now be cheaper on ordinary short-context traffic. V4 Pro must justify its higher bill through a better task success rate.
New DeepSeek V4 API Prices
USD per million tokens:
| Model and period | Cached input | Cache-miss input | Output |
|---|---|---|---|
| DeepSeek V4 Flash, off-peak | $0.007 | $0.22 | $0.66 |
| DeepSeek V4 Flash, peak | $0.014 | $0.44 | $1.32 |
| DeepSeek V4 Pro, off-peak | $0.022 | $0.66 | $1.98 |
| DeepSeek V4 Pro, peak | $0.044 | $1.32 | $3.96 |
Both endpoints retain a 1M-token context window and 384K maximum output. deepseek-v4-flash serves DeepSeek-V4-Flash-0731; deepseek-v4-pro now serves DeepSeek-V4-Pro-0813.
The phrase “2x peak pricing” does not mean the old price simply doubled. Relative to the previous rates, Flash off-peak cache-miss input increased 57%, while off-peak output increased 136%. Peak output is 4.71 times the former output price. Pro’s cached-input change is larger still.
Same Workload, New Bill
For 2M cache-miss input tokens plus 500K output tokens across multiple requests:
| Model | Cost |
|---|---|
| Mistral Small 4 | $0.60 |
| DeepSeek V4 Flash, off-peak | $0.77 |
| GPT-5.6 Luna | $1.00 |
| DeepSeek V4 Flash, peak | $1.54 |
| DeepSeek V4 Pro, off-peak | $2.31 |
| Gemini 3.7 Flash, 2026 introductory price | $3.38 |
| DeepSeek V4 Pro, peak | $4.62 |
| GLM-5.2 | $5.00 |
This is token math, not a quality ranking. Mistral Small 4 has a much smaller context window, while DeepSeek offers 1M context and unusually large output. Gemini adds native image, video, and audio input plus Google tools. Those differences can matter more than a few cents.
Where DeepSeek Still Wins
Off-peak agent scheduling
Off-peak V4 Flash costs 23% less than GPT-5.6 Luna in the example above. If a queue can avoid the seven daily peak hours, scheduling is now a first-class cost control rather than a minor optimization.
Repeated long context
DeepSeek’s off-peak cached input remains inexpensive at $0.007/M for Flash. Stable system prompts, repository summaries, and conversation prefixes can materially reduce the bill, although cache hits are best effort and should be measured from actual usage data.
Long input and long output
V4 keeps a 1M context window and 384K maximum output. A cheaper short-context model is not a direct substitute when a job needs those limits.
Where DeepSeek Lost Its Old Advantage
- Short, uncached traffic: Mistral Small 4 is cheaper in the normalized example.
- Peak-hour traffic: V4 Flash can cost more than GPT-5.6 Luna for the same token mix.
- V4 Pro default routing: At $1.32/$3.96 peak, Pro is no longer a budget route. Escalate to it only when evaluations show fewer failures or retries.
- Multimodal workloads: DeepSeek V4 remains text-first; Gemini 3.7 Flash can absorb text, images, video, audio, and PDFs in one model.
Recommended Routing Policy
| Workload | Starting route |
|---|---|
| Short, routine text | Mistral Small 4 or GPT-5.6 Luna |
| Cache-heavy 1M-context text agent, off-peak | DeepSeek V4 Flash |
| Same agent during DeepSeek peak hours | Compare Luna and Flash using live token ratios |
| Hard DeepSeek-compatible task | V4 Pro after Flash fails evaluation gates |
| Multimodal or Google-grounded task | Gemini 3.7 Flash |
Do not migrate based on unit price alone. Track cost per successful task, total reasoning and output tokens, retries, cache-hit rate, latency, and human repair time.
Bottom Line
DeepSeek still has value, but the value proposition changed from “always the cheapest” to “a strong scheduled, cache-aware long-context route.” V4 Flash is the defensible default within the family. V4 Pro should be an evaluated escalation tier, not a blanket replacement.
Use the AI Pricing Calculator for the conservative peak-rate estimate, and read the current DeepSeek API pricing guide for the full time schedule.
Official sources checked August 17, 2026: DeepSeek pricing, DeepSeek V4 Pro 0813 announcement, Mistral API pricing, OpenAI pricing, and Gemini pricing.
Related Posts
The August 2026 AI API Price War: Google Discounts as DeepSeek Raises Prices
2026-08-01
GPT-5.5 in Codex Pricing (Superseded by GPT-5.6)
2026-04-28
GLM-5.3 vs DeepSeek V4 vs Kimi K3: API Availability, Price, and Agent Fit
2026-08-17