The Known Good Updated 29 Jul 2026
DeepSeek: DeepSeek V3 vs DeepSeek: DeepSeek V4 Pro ·DeepSeek vs DeepSeek

DeepSeek: DeepSeek V3 vs DeepSeek: DeepSeek V4 Pro — head-to-head comparison

DeepSeek: DeepSeek V3 profile DeepSeek: DeepSeek V4 Pro profile All comparisons
DeepSeek: DeepSeek V3 — DeepSeek
Quality
#40 of 62
62based on 4 of 7 evaluations
Known Good Index
Speed
#285 of 317
19.3
Output tokens per second
Price
#110 of 370
$0.35
Blended per 1M tokens
In $0.20 · Out $0.80
Cache hit price
#11 of 238
$0.003
per 1M tokens (−99%)
Context
#255 of 375
164k
tokens (~300 pages)
DeepSeek: DeepSeek V4 Pro — DeepSeek
Quality
#6 of 62
93based on 3 of 7 evaluations
Known Good Index
Speed
#184 of 317
43.7
Output tokens per second
Price
#150 of 370
$0.54
Blended per 1M tokens
In $0.43 · Out $0.87
Cache hit price
#13 of 238
$0.004
per 1M tokens (−99%)
Context
#25 of 375
1.05M
tokens (~1,600 pages)
Summary

On the Known Good Index, DeepSeek: DeepSeek V3 scores 62 and DeepSeek: DeepSeek V4 Pro scores 93 — a 31-point lead for DeepSeek: DeepSeek V4 Pro. They share 2 published evaluations, with DeepSeek: DeepSeek V3 ahead on 0 and DeepSeek: DeepSeek V4 Pro ahead on 2.

Blended pricing is $0.35 per 1M tokens for DeepSeek: DeepSeek V3 and $0.54 for DeepSeek: DeepSeek V4 Pro — DeepSeek: DeepSeek V3 is about 1.6× cheaper.

Throughput is 19 tokens per second for DeepSeek: DeepSeek V3 and 44 for DeepSeek: DeepSeek V4 Pro, with time to first token at 1.4s and 1.9s. DeepSeek: DeepSeek V3 accepts 164k tokens of context against 1.05M.

DeepSeek: DeepSeek V3 is open weights under Open weights (unrestricted); DeepSeek: DeepSeek V4 Pro is open weights under Open weights (unrestricted). 19 tracked hosts serve one or both, and 3 serve both.

Generated from tracked data on 29 Jul 2026 · how these comparisons are computed →

At a glance
MetricDeepSeek: DeepSeek V3DeepSeek: DeepSeek V4 Pro
Known Good Index 62 based on 4 of 7 evaluations 93 based on 3 of 7 evaluations
Blended $/1M $0.35 $0.54
Output tok/s 19 44
Context window 164k 1.05M
Providers 3 19
Jump to section

Quality

Known Good Index and the capability indices, side by side

Index scores

Each index is min-max normalised across tracked models and scaled 0–100 from published evaluations. We do not run these evaluations.

MetricDeepSeek: DeepSeek V3DeepSeek: DeepSeek V4 Pro
Known Good Index 62 93
Mathematics Index 57 97
Reasoning Index 61 —
Agentic Index — 89
Openness Index 60 60
The Known Good

Quality Breakdown

Every evaluation both models report a published score for

Evaluations both models report

Bar = DeepSeek: DeepSeek V3 · marker = DeepSeek: DeepSeek V4 Pro · scores ingested from Epoch AI (CC-BY 4.0)

GPQA Diamond71 / 90
Mock AIME 2024–2549 / 97

Right-hand figures read DeepSeek: DeepSeek V3 / DeepSeek: DeepSeek V4 Pro. DeepSeek: DeepSeek V3 leads on 0 of 2, DeepSeek: DeepSeek V4 Pro on 2.

The Known Good

Arena Elo

Human preference rating from blind pairwise votes

Arena ratings

Elo from blind pairwise votes, ingested from Arena. Higher is better.

MetricDeepSeek: DeepSeek V3DeepSeek: DeepSeek V4 Pro
Text Arena 1,333 1,450
Text Arena (style controlled) 1,359 1,457
The Known Good

Price & Cost

Published API pricing by token type

Published pricing

USD per 1M tokens. Blended is a 3:1 input:output mix. Live from OpenRouter, cross-checked against models.dev. Lower is better.

MetricDeepSeek: DeepSeek V3DeepSeek: DeepSeek V4 Pro
Input $/1M $0.20 $0.43
Output $/1M $0.80 $0.87
Blended $/1M $0.35 $0.54
Cache read $/1M $0.003 $0.004
The Known Good
We publish blended price rather than cost-per-task. Cost-per-task requires token counts from an evaluation suite we do not run. Blended price is computed from published rates and is fully reproducible — see the pricing methodology.

Context Window

Maximum input accepted and maximum output emitted

Context and output limits

Provider-declared limits, from metadata via models.dev and OpenRouter. Higher is better.

MetricDeepSeek: DeepSeek V3DeepSeek: DeepSeek V4 Pro
Context window 164k 1.05M
Max output tokens 16k 384k
The Known Good

Speed & Latency

Median throughput and time to first token across hosting providers

Output speedTime to first token
Throughput and latency

Medians across all tracked providers, derived from OpenRouter endpoint telemetry.

MetricDeepSeek: DeepSeek V3DeepSeek: DeepSeek V4 Pro
Output tokens/s 19 44
Time to first token 1.4s 1.9s
The Known Good

Providers

Which hosts serve each model, and at what price and speed

19 hosts · 3 serve both
Full provider benchmarks →
Provider DeepSeek: DeepSeek V3 blendedDeepSeek: DeepSeek V3 tok/s DeepSeek: DeepSeek V4 Pro blendedDeepSeek: DeepSeek V4 Pro tok/s
StreamLake $0.35 21 $0.84 38
Novita $0.62 20 $1.46 62
DeepInfra $0.46 12 $1.62 51
BaseTen — — $2.17 110
Parasail — — $2.17 65
Cloudflare — — $2.17 64
Venice — — $2.06 55
Fireworks — — $2.17 53
Alibaba · first party — — $2.01 50
GMICloud — — $0.85 48
Together — — $2.17 45
SiliconFlow — — $1.91 41
CoreWeave — — $2.17 40
DeepSeek · first party — — $0.54 40
Baidu · first party — — $0.78 38
Wafer — — $1.50 28
AtlasCloud — — $2.10 25
DigitalOcean — — $1.74 8
Ionstream — — $1.41 5

Per-host pricing and telemetry from OpenRouter. A dash means that host does not serve that model.

The Known Good
Sources for this page
Pricing & providers — OpenRouter Metadata — models.dev Benchmarks — Epoch AI (CC-BY 4.0)Elo — Arena Full attribution →