---
title: "DeepSeek V3 685B Benchmarks - RankedAGI"
description: "Detailed benchmark and metadata record for DeepSeek V3."
source: "https://rankedagi.com/models/deepseek-3"
---

# DeepSeek V3

| Field | Value |
| --- | --- |
| Organization | DeepSeek |
| License | Open Source |
| Size | 685B |
| Version | Latest |
| Released | 2025-03-24 |
| Context window | 128K |
| Knowledge cutoff | 2024-07-01 |
| Input cost per million tokens | $0.27 |
| Output cost per million tokens | $1.1 |
| Cached input cost per million tokens | $0.07 |
| Model ID | deepseek-v3-0324 |
| Last updated | 2026-05-04T03:43:08.354Z |

## Benchmarks

| Benchmark | Category | Value | Description | Source |
| --- | --- | --- | --- | --- |
| RankedAGI Coding | coding | 40.5% | RankedAGI Coding Score |  |
| RankedAGI Agentic | Agents | 27.8% | RankedAGI Agentic Score |  |
| LiveCodeBench v6 | coding | 29.0% | LiveCodeBench v6 - Evaluation of Large Language Models for Code |  |
| LiveCodeBench v5 | coding | 49.2% | LiveCodeBench - Evaluation of Large Language Models for Code |  |
| ChatArena (LMSYS) | coding | 1343 | ChatArena (LMSYS) Coding ELO Score with Style Control |  |
| CyberGym | safety | 3.6% |  |  |
| Aider Polyglot | coding | 55.1% | Aider Polyglot Code Completion Benchmark |  |
| LiveBench Coding Score (old) | coding | 70.9% | LiveBench Coding Score old version |  |
| RankedAGI Reasoning | reasoning | 40.1% | RankedAGI Reasoning Score |  |
| GPQA Diamond | reasoning | 68.4% | Generalized Prefix Question Answering Score (Reasoning) PhD Level Reasoning |  |
| Text Arena | general | 1395 | ChatArena (LMSYS) ELO Score | https://arena.ai/leaderboard/text |
| AIME 2024 | math | 59.4% | AIME 2024 Competition Math |  |
| NYT Connections Extended | reasoning | 17.4% | NYT Connections Extended Version | https://github.com/lechmazur/nyt-connections/ |
| MMLU Pro | knowledge | 81.2% | Graduate-level visual problem-solving |  |
| Hughes Hallucination Evaluation | safety | 8.0% | Computed using Vectara's Hallucination Evaluation Model, also known as HHEM. This evaluates how often an LLM introduces hallucinations when summarizing a document. | https://github.com/vectara/hallucination-leaderboard |
| LiveBench Average | general | 66.9% | LiveBench Average Score old version |  |
| LiveBench Coding 25.4 | coding | 40.5% | LiveBench Coding Score 2025-04 | https://livebench.ai/ |
| LiveBench Data Analysis | general | 60.4% | LiveBench Data Analysis Score | https://livebench.ai/ |
| LiveBench Language | general | 49.1% | LiveBench Language Score | https://livebench.ai/ |
| RankedAGI Math | math | 54.9% | RankedAGI Math Score |  |
| RankedAGI Overall | general | 41.8% | Overall RankedAGI score |  |
| GDPval-AA Elo | Agents | 407 | Office Tasks (Artificial Analysis) |  |
| Code DesignArena | Design | 1159 |  | https://www.designarena.ai/leaderboard/code |

## Model Links

| Title | URL |
| --- | --- |
| Release | https://huggingface.co/deepseek-ai/DeepSeek-V3-0324 |