Skip to content
llmwaves

DeepSeek V3

DeepSeek · released Dec 26, 2024

Open weightsTool useStructured output

DeepSeek-V3 is the latest model from the DeepSeek team, building upon the instruction following and coding abilities of the previous versions. Pre-trained on nearly 15 trillion tokens, the reported evaluations...

Specification

Context window
164K
Max output
16K
Knowledge cutoff
Jul 31, 2024
Parameters
684.5B
Licence
Proprietary
Serving providers
3
Moderated
No
Uptime
100.0%

Intelligence

Not benchmarked

Coding

Coding Index

Agentic

Agentic Index

Output speed

27 t/s

Median across providers

Latency

428ms

Time to first token

Cost per task

$0.04

Estimated

Benchmarks

Where the score comes from

The Intelligence Index is a composite. These are the underlying evaluations this model was actually measured on.

Evaluation scores

This model has not been put through the benchmark suite. Its speed, latency and pricing figures below are still measured.

Against its peers

Intelligence Index · this model highlighted, nearest peers in grey

Peers are the models sitting closest on the Intelligence Index — the set you would realistically choose between.

View as table
ModelIntelligence
Claude Opus 563.1
Claude Fable 562.1
GPT-5.6 Sol60.9
Kimi K359.7
Qwen3.8 Max58.1
Claude Opus 4.857.3
Muse Spark 1.256.8

Percentile among all indexed models

Arena Elo37th

Pricing

What it costs to run

List prices per million tokens, plus what one representative task works out to.

List price

Input / 1M tokens
$0.257
Output / 1M tokens
$1.03
Cached input / 1M
Not offered
Blended 3:1
$0.45

One task, estimated

$0.04

Input tokens
50,000
Output tokens
25,000
Profile
Standard

Estimated from list pricing: 50K input tokens plus 80K output tokens for reasoning models (25K for non-reasoning).

Serving providers

Speed and latency figures are medians across these providers, so a widely-served model reports a blend rather than any single endpoint.

Arena

Head-to-head generation quality

Elo from pairwise judgements, broken out by the kind of thing the model was asked to build.

Overall Elo
1,143
Win rate
48.5%
Strongest at
Websites
Tournaments
14,721

Elo by category

Dot position on a 1,000–1,150 scale · Elo has no meaningful zero

Agent categories (full-stack apps, mobile apps) are only scored for models tested in agent harnesses.

View as table
CategoryEloWin rate
Websites114348.5%
3D scenes114250.6%
UI components112852.8%
Data visualisation112351.2%
Game development109543.9%
SVG102338.8%