# DeepSeek V4 Pro

DeepSeek, released 2026-07-19

**Verdict: recommended.** Strong price-performance on open MIT weights; the 2x peak window is 01:00-04:00 and 06:00-10:00 UTC on weekdays only, so weekends bill entirely off-peak. Last verified 2026-07-24.

## Pricing

- Input: $0.66 / 1M tokens
- Output: $1.98 / 1M tokens
- Context: 1M tokens; 2x price at weekday peak hours

## Strengths

- Cost-sensitive coding and autonomous agent workloads: far cheaper per session than GPT-5.5 or Opus 4.8 at competitive quality, even after the August 2026 price rise
- Self-hosted enterprise and regulated-industry deployment: MIT license means proprietary code never leaves your infrastructure
- Competitive programming and code-generation benchmarks: 93.5% LiveCodeBench, 3206 Codeforces Elo
- Weekend and overnight batch work: peak billing applies only Monday to Friday, so all weekend hours bill at the off-peak $0.66 / $1.98 rate
- Terminal and CLI agent tasks: trails GPT-5.5 (67.9% vs 75.1% on Terminal-Bench 2.0)
- Workloads requiring Western data-residency guarantees without self-hosting: DeepSeek's API is China-hosted
- Weekday workloads pinned to peak hours: rates double to $1.32 / $3.96 during 01:00-04:00 and 06:00-10:00 UTC, Monday through Friday only

## Benchmarks

- SWE-bench Verified: 80.6% (https://build.nvidia.com/deepseek-ai/deepseek-v4-pro/modelcard)
- LiveCodeBench (Pass@1): 93.5% (https://huggingface.co/blog/ResterChed/deepseek-v4-ga-architecture)
- Codeforces Elo: 3206 (https://huggingface.co/blog/ResterChed/deepseek-v4-ga-architecture)
- Terminal-Bench 2.0: 67.9% (https://huggingface.co/blog/ResterChed/deepseek-v4-ga-architecture)
- SWE Pro (Resolved): 55.4% (https://build.nvidia.com/deepseek-ai/deepseek-v4-pro/modelcard)
- MCPAtlas Public: 73.6% (https://huggingface.co/blog/ResterChed/deepseek-v4-ga-architecture)
- GPQA Diamond: 71.7 (https://pricepertoken.com/pricing-page/model/deepseek-deepseek-v4-pro)

## Sources

- [MarkTechPost — Kimi K3 vs V4 Pro vs GLM-5.2](https://www.marktechpost.com/2026/07/18/kimi-k3-vs-deepseek-v4-pro-vs-glm-5-2-open-trillion-scale-moe-models-compared-on-benchmarks-license-and-serving-cost/)
- [BenchLM — DeepSeek API Pricing: V4 Pro & Flash Rates](https://benchlm.ai/deepseek/api-pricing)
- [Tech Insider — Best AI Model for Coding 2026](https://tech-insider.org/best-ai-model-for-coding-2026/)
- [DeepSeek V4 GA Architecture (Hugging Face)](https://huggingface.co/blog/ResterChed/deepseek-v4-ga-architecture)
- [NVIDIA Build — deepseek-v4-pro Model Card](https://build.nvidia.com/deepseek-ai/deepseek-v4-pro/modelcard)
- [PricePerToken — DeepSeek V4 Pro Pricing](https://pricepertoken.com/pricing-page/model/deepseek-deepseek-v4-pro)
- [DeepSeek API Docs: Models and Pricing](https://api-docs.deepseek.com/quick_start/pricing/)

Canonical: https://neomanex.com/models/deepseek-v4-pro