What the evidence says
- Developer
- DeepSeek
- Released
- December 24, 2024
- Domains
- Language
- Tasks
- Language modeling/generation · Code generation · Quantitative reasoning · Question answering
- Access
- Open weights (restricted use)
- Weights
- Open weights
Across the boards
Each row is one named authority's own current view. Nothing is combined.
- LMArena overall
- 1358.5 · rank #172 · 21,770 battles · published July 20, 2026 · board
- LMArena coding arena
- 1387.6 · rank #187 · board
- LMArena creative writing arena
- 1348.8 · rank #146 · board
- LMArena instruction following arena
- 1343.8 · rank #176 · board
- LMArena hard prompts arena
- 1350.3 · rank #198 · board
- LMArena multi-turn arena
- 1373.6 · rank #156 · board
- LMArena long queries arena
- 1374.7 · rank #162 · board
- LMArena math arena
- 1310.7 · rank #203 · board
Benchmark results
Epoch AI Benchmarking Hub scores for this reviewed identity. Rows with a scaffold are system results, not bare-model results. Fraction-unit scores shown as percentages.
- Aider Polyglot
- 48.4% · diff
- FrontierMath
- 1.7%
- GPQA Diamond
- 56.5%
- MATH Level 5
- 64.9%
- OTIS Mock AIME 2024-2025
- 15.8%
Common questions
What is DeepSeek-V3?
DeepSeek-V3 is a DeepSeek model released December 24, 2024, according to Epoch AI's reviewed catalog record. The record classifies it under language.
How does DeepSeek-V3 rank on LMArena?
In the LMArena style-controlled text snapshot published July 20, 2026, DeepSeek-V3 holds official rank 172 with a rating of 1358.5 across 21,770 battles.
What does DeepSeek-V3 cost?
No reviewed source carries pricing for this record, so no price is reported here.