TL;DR
This dataset aggregates API pricing and core specs for three DeepSeek V4 models (V4 Pro / V4 Flash / V4.1 Flash), with prices listed for peak and off-peak tiers as of September 10, 2026, when V4.1 Flash pricing took effect.
Data notes
- Scope: 3 DeepSeek V4 models (V4 Pro, V4 Flash, V4.1 Flash).
- Pricing unit: per 1M tokens; peak hours are Mon–Fri 9:00–12:00 and 14:00–18:00 Beijing time; CNY prices are official announcement prices, USD prices from the official pricing page (as compiled by Apidog).
- As of: September 10, 2026 (effective date of V4.1 Flash pricing).
- Status: V4 Flash retired, requests served by V4.1 Flash; V4 Pro retires at 12:00 Beijing time on September 14, 2026, rerouted to V4.1 Flash.
Core data
Table 1: DeepSeek V4 series API pricing (USD / per 1M tokens)
| model | price_tier | input_cache_hit_usd | input_cache_miss_usd | output_usd |
|---|---|---|---|---|
| V4 Pro | off_peak | 0.022 | 0.66 | 1.98 |
| V4 Pro | peak | 0.044 | 1.32 | 3.96 |
| V4 Flash (old) | off_peak | 0.007 | 0.22 | 0.66 |
| V4 Flash (old) | peak | 0.014 | 0.44 | 1.32 |
| V4.1 Flash | off_peak | 0.003 | 0.15 | 0.60 |
| V4.1 Flash | peak | 0.006 | 0.30 | 1.20 |
Table 2: DeepSeek V4.1 Flash official CNY pricing (per 1M tokens)
| model | price_tier | input_cache_hit_cny | input_cache_miss_cny | output_cny |
|---|---|---|---|---|
| V4.1 Flash | off_peak | 0.02 | 1 | 4 |
| V4.1 Flash | peak | 0.04 | 2 | 8 |
Table 3: DeepSeek V4 series core specs
| model | architecture | total_params | active_params | context_tokens | max_output_tokens | vision | open_source | concurrency_limit |
|---|---|---|---|---|---|---|---|---|
| V4 Pro | V4 MoE | ~1.6T | ~49B | 1M | 384K | Not disclosed | Not disclosed | 500 |
| V4 Flash (old) | V4 MoE | ~284B | ~13B | 1M | 384K | Separate Vision-Exp build | Not disclosed | 2500 |
| V4.1 Flash | V4.1 Causal Encoder-Decoder (40 layers) | 552B | 8B input / 16B output | 1M | 384K | Native (DeepSeek-ViT) | MIT (weights open) | 2500 |
Field notes
- model: model name.
- price_tier: tier; peak is Mon–Fri 9:00–12:00 and 14:00–18:00 Beijing time; off_peak is the rest.
- input_cache_hit_usd / input_cache_miss_usd / output_usd: per-1M-token input (cache hit/miss) and output prices in USD.
- input_cache_hit_cny / input_cache_miss_cny / output_cny: per-1M-token input (cache hit/miss) and output prices in CNY.
- total_params: total parameter count; active_params: parameters activated per inference.
- context_tokens: context window length; max_output_tokens: maximum output length per request.
- concurrency_limit: API concurrency cap.