DeepSeek V3 pricing for code generation

Quick start

Call DeepSeek V3 with the OpenAI Python SDK pointed at api.deepseek.com:

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://api.deepseek.com/v1",
    api_key=os.environ["DEEPSEEK_API_KEY"],
)
response = client.chat.completions.create(
    model="deepseek-v3",
    max_tokens=4000,
    messages=[{"role": "user", "content": "Your prompt here"}]
)
# Cost: ~$0.004940 per call at standard rates
# Input:  $0.27/MTok
# Output: $1.10/MTok

How DeepSeek V3 compares on cost

Per-call cost for all models scoring 50+ on code generation benchmarks. Hover for details.

Highest-quality affordable option: Claude Haiku 4.5 at $0.0220/call (4.5x the cost, 72/100 quality). DeepSeek V3 is already the cheapest choice for code generation.

What drives code generation cost

Cost here is dominated by output, not input: the model writes far more than you ask. A prompt with a short spec and a few reference files runs 2K input, but a complete module comes back at 4K output or more, so a model's output rate matters roughly twice as much as its input rate. The biggest saving is not a cheaper model: it is caching the system prompt and coding standards you resend on every call.

For DeepSeek V3 specifically, that means output tokens cost 4.1x more than input, at $1.10 vs $0.27 per MTok. At this use case’s 2,000/4,000 token split, output accounts for 89% of the $0.004940 per-call cost.

DeepSeek V3 quality benchmarks for code generation

Benchmark data for code generation is pending for DeepSeek V3. Test against your own evaluation suite before moving to production.

Benchmark data is not yet available for DeepSeek V3. We recommend testing against your own evaluation criteria.

About DeepSeek V3

Prices are off-peak rates from api-docs.deepseek.com (read 2026-10-11). Peak-hour prices are 2x the listed input and output rates — Tokeven does not model time-of-day tiers; off-peak is used as the base. Cache hit: $0.07/MTok input. No batch discount published. Not yet benchmarked, so never recommended.

Other models from DeepSeek

You can compare DeepSeek V3 head-to-head against any other model on our model comparison pages, or track how pricing has changed over time on the pricing history tracker.

Frequently asked questions