DeepSeek

DeepSeek V3

Type
language model
Context
164K tokens
Max output
8K
Released
26 December 2024
API string
deepseek-v3
Weights
open

Prices by provider

Provider Input, $ per 1M Output, $ per 1M In our data since
Hyperbolic $0.2 $0.2 31 Jul 2026
siliconflow-cn $0.25 $1 31 Jul 2026
SiliconFlow $0.25 $1 31 Jul 2026
DeepSeek $0.27 $1.1 31 Jul 2026
Vercel $0.27 $1.12 31 Jul 2026
drun $0.28 $1.1 31 Jul 2026
alibaba-cn $0.287 $1.147 31 Jul 2026
DeepInfra $0.38 $0.89 31 Jul 2026
Nebius Token Factory $0.5 $1.5 31 Jul 2026
Helicone $0.56 $1.68 31 Jul 2026
vercel_ai_gateway $0.9 $0.9 31 Jul 2026
Fireworks AI $0.9 $0.9 31 Jul 2026
Microsoft Foundry $1.14 $4.56 31 Jul 2026
Together AI $1.25 $1.25 31 Jul 2026
Replicate $1.45 $1.45 31 Jul 2026

Only the base tier and only per-token prices are shown. Batch, discounted and cached rates, as well as prices per image or per second of video, do not go into this table: they cannot stand in the same column as a price per million tokens.

The date in the last column is the day this price first entered our collection. The price may well be older: before that day we simply were not recording it. It has not changed since — otherwise a new row with a new date would stand in its place.

Measurement results

Task set Result Run conditions Measured by
Arena Score in Spanish 1,350.04 1,302–1,398
Arena Score, multi-turn dialogue 1,349.12 1,339–1,359
Arena Score in English 1,345.89 1,340–1,352
Arena Score, long queries 1,343.32 1,333–1,354
Arena Score in French 1,341.9 1,304–1,379
Arena Score, overall 1,332.64 1,328–1,337
Arena Score, creative writing 1,329.44 1,319–1,339
Arena Score, programming 1,325.47 1,315–1,336
Arena Score in Russian 1,323.22 1,312–1,335
Arena Score, instruction following 1,315.51 1,309–1,322
Arena Score, hard prompts 1,312.24 1,304–1,320
Arena Score, mathematics 1,310.67 1,300–1,321
Arena Score, expert questions 1,305.19 1,289–1,322
MATH, difficulty level five 64.85 % · with a tuned harness Epoch evaluations
Aider Polyglot — code edits in six languages 48.4 % · with a tuned harness Aider LLM Leaderboards
GPQA Diamond — graduate-level questions 42.05 % · with a tuned harness Epoch evaluations
Mock AIME 2024–2025 — olympiad problems 15.75 % · with a tuned harness Epoch evaluations
SimpleBench — trick questions 2.68 % SimpleBench Leaderboard
CritPt — physics problems 0 %