DeepSeek-V3: Pricing, Context, and Benchmarks

DeepSeek-V3 is DeepSeek's DeepSeek-V3 model released on 2024-12-26. It runs on a 128K-token context window, priced at $0.27/M input and $1.10/M output. 671B MoE with 37B active. GPT-4-class quality at ~10x lower cost. Aggressive cache discounts.

Specifications

Benchmark scores

Scores reflect single measurements on the date shown. Reproducibility varies; consult the linked source for methodology. Use the comparison tool for side-by-side analysis.

What this model is good for

Based on benchmark scores and capability fit, DeepSeek-V3 ranks well for:

• Best Open-Weight Model — rank #1. Best overall open-weight quality at the lowest token cost.
• High-Throughput Inference — rank #2. Fast on Groq and Together. Cheaper per million tokens.
• Code Generation — rank #5. Excellent price/performance. Use cached input pricing aggressively.

Cost calculator

For 1,000 tokens of input and 1,000 tokens of output per call, this model costs $0.0014 per call. For typical usage:

For more precise estimates including caching discounts, use the LLM API Cost Calculator.

Other DeepSeek models

• DeepSeek-R1

Common comparisons

• DeepSeek-V3 vs GPT-5 Mini Side-by-side specs, pricing, and benchmarks
• DeepSeek-V3 vs Gemini 2.5 Flash Side-by-side specs, pricing, and benchmarks
• DeepSeek-V3 vs Codestral Side-by-side specs, pricing, and benchmarks
• DeepSeek-V3 vs GPT-4o Mini Side-by-side specs, pricing, and benchmarks

API setup

curl https://api.deepseek.com/v1/chat/completions \
  -H "Authorization: Bearer $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "deepseek-v3", "messages": [{"role": "user", "content": "Hello"}]}'

See the official documentation for the full request/response schema. For pre-flight token counting before paying, use the appropriate token counter.