DeepSeek V4 API, Pricing, Benchmarks & Guides
Follow DeepSeek V4 news, compare it with ChatGPT, Claude, Gemini, and Grok, read Reasonix and local deployment guides, and jump to in-stock one-off Coding Plans when you need configured API access.
- Models
- deepseek-v4-pro · v4-flash
- Context
- 1M tokens
- API
- OpenAI-compatible
- Off-peak from
- $0.007 / 1M cache-hit input
DeepSeek V4 Chat, Harness and Developer Guides
Evidence-first benchmark update
DeepSeek V4 Pro-0813: official agent benchmark update and frontier comparison
DeepSeek's Pro-0813 release table is shown separately from the independent panel so scores produced with different test harnesses are not presented as one controlled ranking.
DeepSeek release status
- V4 Flash-0731
- Official API release · Public Beta · Version 0731
- V4 Pro-0813
- Official API model · Version 0813
Independent comparison: one shared methodology
Artificial Analysis Intelligence Index v4.1 · maximum-effort model variants. Scores in this panel are directly comparable with each other.
Intelligence Index · 0–70
Claude Fable 5
Adaptive reasoning · Max effort · Opus 4.8 fallback
Source60GPT-5.6 Sol
Max effort
Source59Claude Opus 4.8
Max effort
Source56
DeepSeek V4 Pro-0813
Separate release methodologyThe release-table scores below are not substituted into the independent Intelligence Index panel above.
DeepSeek V4 Pro-0813 release benchmarks
DeepSeek's August 13 comparison reports Pro-0813 alongside Flash-0731, both previews, Opus-4.8, and Fable 5 across Terminal Bench 2.1 and eight other agent tasks.
Terminal Bench 2.1
Scores from DeepSeek's release comparison table. Higher scores indicate stronger reported terminal-agent performance.
| Model | Reported score |
|---|---|
| DeepSeek-V4-Pro-0813 | 87.9 |
| DeepSeek-V4-Flash-0731 | 82.7 |
| DeepSeek-V4-Pro-Preview | 72.1 |
| DeepSeek-V4-Flash-Preview | 61.8 |
| Opus-4.8 | 85.0 |
| Fable 5 (w/ fallback) | 88.0 |
Six-model comparison across nine code-agent benchmarks
Higher bars indicate higher reported scores. Fable 5 has no reported result for NL2Repo or Agents' Last Exam.
Reported score · 0–100
| Benchmark | DeepSeek-V4-Pro-0813 | DeepSeek-V4-Flash-0731 | DeepSeek-V4-Pro-Preview | DeepSeek-V4-Flash-Preview | Opus-4.8 | Fable 5 (w/ fallback) |
|---|---|---|---|---|---|---|
| Terminal Bench 2.1 | 87.9 | 82.7 | 72.1 | 61.8 | 85.0 | 88.0 |
| NL2Repo | 61.5 | 54.2 | 38.5 | 39.4 | 69.7 | Not reported |
| CyberGym | 83.3 | 76.7 | 52.7 | 38.7 | 78.3 | 83.1 |
| DeepSWE | 62.7 | 54.4 | 12.8 | 7.3 | 58.0 | 70.0 |
| Toolathlon-Verified | 74.1 | 70.3 | 55.9 | 49.7 | 76.2 | 77.9 |
| Agents' Last Exam | 25.7 | 25.2 | 16.5 | 15.8 | 25.7 | Not reported |
| AutomationBench (Public) | 31.8 | 25.1 | 12.8 | 10.8 | 27.2 | 29.1 |
| DSBench-FullStack | 71.1 | 68.7 | 41.8 | 37.0 | 71.6 | 77.2 |
| DSBench-Hard | 67.2 | 59.6 | 31.1 | 25.8 | 71.7 | 68.3 |
Benchmark data: DeepSeek V4 Pro-0813 release comparison table · August 13, 2026 · Confirm the current Pro-0813 API model version
DeepSeek V4 Flash and Pro API token pricing
Official peak/off-peak rates per 1M tokens. Peak hours are 09:00–12:00 and 14:00–18:00 Beijing time (01:00–04:00 and 06:00–10:00 UTC). These rates are separate from this site's Coding Plans.
| Model | Cache hit input | Cache miss input | Output | Pricing window |
|---|---|---|---|---|
| DeepSeek V4 Flash | $0.007 / ¥0.05 | $0.22 / ¥1.50 | $0.66 / ¥4.50 | Off-peak · from August 17 |
| DeepSeek V4 Flash | $0.014 / ¥0.10 | $0.44 / ¥3.00 | $1.32 / ¥9.00 | Peak · from August 17 |
| DeepSeek V4 Pro | $0.022 / ¥0.15 | $0.66 / ¥4.50 | $1.98 / ¥13.50 | Off-peak · from August 17 |
| DeepSeek V4 Pro | $0.044 / ¥0.30 | $1.32 / ¥9.00 | $3.96 / ¥27.00 | Peak · from August 17 |
Last verified: August 13, 2026
DeepSeek's official model documentation now identifies deepseek-v4-pro as version Pro-0813. Benchmark coverage remains separate from Coding Plan inventory and availability.
Four Coding Plans · One-time or monthly
DeepSeek Coding Plan leads the current lineup
Get 50M tokens every 5 hours and 400M tokens every 7 days across DeepSeek V4.1 Flash and the existing V4 routes. New configurations use the native multimodal Model ID deepseek-flash. Also in stock while inventory lasts: Agent Plan, Coding Plan Plus, and Coding Plan Max — each with one-time purchase or 50%-off monthly renewal.