The Known Good Updated 25 Jul 2026
DeepSeek ·Open weights (unrestricted) ·Released April 2026·Tracked since 25 Jul 2026

DeepSeek: DeepSeek V4 Pro — quality, speed & price analysis

Compare with another model Provider benchmarks Data sources
Quality
#6 of 61
93based on 3 of 6 evaluations
Known Good Index
Speed
#182 of 316
45.2
Output tokens per second
Price
#145 of 340
$0.54
Blended per 1M tokens
In $0.43 · Out $0.87
Cache hit price
#12 of 206
$0.004
per 1M tokens (−99%)
Context
#25 of 345
1.05M
tokens (~1,600 pages)
Summary

DeepSeek: DeepSeek V4 Pro is in the top quarter of the models we track for quality, and inexpensive for its standing, and notably slow. It accepts text input, outputs text, and supports a 1.05M-token context window.

It scores 93 on the Known Good Index, well above the tracked average of 60, from 3 of the six component evaluations. Measured against the tracked average, its strongest result is MATH / AIME (97 against 53) and its weakest is SWE-bench Verified (78 against 71).

Pricing is $0.43 per 1M input tokens (below the $2.18 average) and $0.87 per 1M output (below the $9.91 average). Cached input reads at $0.004 per 1M, a 99% discount on its input price. At 45 tokens per second it sits in the lower middle of the tracked range.

Generated from tracked data on 25 Jul 2026 · how these comparisons are computed →

Technical specifications
ReasoningYes
Input modalitytext
Output modalitytext
Context window1,048,576 tokens
Max output384,000 tokens
WeightsOpen weights · Open weights (unrestricted)
Knowledge cutoffMay 2025
Tool callingYes
Providers18 hosts · compare →
Highlights — where DeepSeek: DeepSeek V4 Pro sits
Quality

Known Good Index · Higher is better · Composite of six published evaluations

98
96
95
94
93
93
93
91
90
90
88
88
83
gpt-5.5-pre-release
Gemini 3.1 Pro
Preview
Gemini 3.5 Flash
Claude Opus 4.7
Qwen3.7 Max
DeepSeek V4 Pro
Kimi K2.6
GLM 5.2
Qwen3.6 Max
Preview
GPT-5.4
Claude Opus 4.6
GLM 5.1
Gemini 3 Flash
Preview
Speed

Output tokens per second · Higher is better · OpenRouter endpoint telemetry

3,894
1,728
1,134
578
449
387
320
179
173
172
168
166
45
Relace Apply 3
Morph V3 Large
Morph V3 Fast
Grok 4.20
Multi-Agent
gpt-oss-safeguard-20b
Nano Banana 2 Lite
(Gemini 3.1 Flash Lite Image)
Mercury 2
Nano Banana
(Gemini 2.5 Flash Image)
Llama 3.2 3B
Instruct
gpt-oss-120b
Laguna XS 2.1
Nano Banana 2
(Gemini 3.1 Flash Image Preview)
DeepSeek V4 Pro
Price

Blended USD per 1M tokens · Lower is better · Published pricing via OpenRouter

$0.01
$0.02
$0.04
$0.04
$0.04
$0.06
$0.06
$0.06
$0.06
$0.06
$0.06
$0.06
$0.06
$0.54
Ling-2.6-flash
Mistral Nemo
Granite 4.0 Micro
Llama 3 8B Lunaris
Nex-N2-Mini
Qwen2.5 7B
Instruct
gpt-oss-20b
Mistral Small 3
Llama 3.1 8B
Instruct
MythoMax 13B
Nova Micro 1.0
Granite 4.1 8B
Gemma 3 4B
DeepSeek V4 Pro
Jump to section

Quality

How DeepSeek: DeepSeek V4 Pro compares against every tracked model

Known Good IndexCoding IndexCapability indicesBy open weights
Known Good Index

Composite of GPQA Diamond, MATH / AIME, SWE-bench Verified, LiveBench, Humanity's Last Exam and Terminal-Bench, normalised 0–100. Scores are ingested from Epoch AI; we do not run these evaluations.

13 of 908 models ▾ ☰⚙⇩ +Add model from specific provider
98
96
95
94
93
93
93
91
90
90
88
88
83
💡 Reasoning models are indicated by a lightbulb
The Known Good

Quality Breakdown

DeepSeek: DeepSeek V4 Pro on each evaluation feeding the index, against the tracked average

Evaluation scores

Bar = DeepSeek: DeepSeek V4 Pro · grey marker = average across tracked models · scores ingested from Epoch AI

GPQA Diamond90 / 63
MATH / AIME97 / 53
SWE-bench Verified78 / 71
The Known Good

Arena Elo

Human preference rating from blind pairwise votes

Text Arena

Elo from blind pairwise votes, ingested from Arena

1449
Elo · 95% CI ±4
#36
of 381 ranked models
45k
votes counted
1457
style-controlled (#46)

Ingested 25 Jul 2026.

The Known Good

Price & Cost

Published API pricing by token type

BlendedInput vs outputCache discountQuality per dollar
Blended Price

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

14 of 908 models ▾ ☰⚙⇩ +Add model from specific provider
$0.01
$0.02
$0.04
$0.04
$0.04
$0.06
$0.06
$0.06
$0.06
$0.06
$0.06
$0.06
$0.06
$0.54
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 tokens accepted

Context windowMax output tokens
Context Window

Maximum input context from provider metadata via models.dev and OpenRouter. Higher is better.

13 of 908 models ▾ ☰⚙⇩ +Add model from specific provider
2.00M
2.00M
2.00M
2.00M
2.00M
1.31M
1.05M
1.05M
1.05M
1.05M
1.05M
1.05M
1.05M
The Known Good

Speed

Median output throughput across hosting providers

Output speedBy provider
Output Speed

Median output tokens per second across all tracked providers. Derived from OpenRouter endpoint telemetry. Higher is better.

13 of 908 models ▾ ☰⚙⇩ +Add model from specific provider
3,894
1,728
1,134
578
449
387
320
179
173
172
168
166
45
The Known Good

Latency

Time to first token, median across providers

Time to first tokenEnd-to-end response
Latency (TTFT)

Seconds to first streamed chunk, median across tracked providers. Derived from OpenRouter endpoint telemetry. Lower is better.

13 of 908 models ▾ ☰⚙⇩ +Add model from specific provider
0.2
0.2
0.2
0.2
0.2
0.2
0.2
0.2
0.2
0.2
0.3
0.3
2.0
The Known Good

Providers

Hosts serving DeepSeek: DeepSeek V4 Pro, with their own price and speed

18 providers
Full provider benchmarks →
ProviderInput $/MOutput $/M BlendedSpeedTTFTContext
BaseTen $1.74 $3.48 $2.17 110 0.7s 262k
Fireworks $1.74 $3.48 $2.17 66 2.5s 1.05M
Novita $1.17 $2.34 $1.46 66 1.5s 1.05M
DeepInfra $1.30 $2.60 $1.62 65 0.5s 1.05M
GMICloud $0.68 $1.36 $0.85 62 2.6s 1.05M
Venice $1.65 $3.30 $2.06 61 1.0s 1.00M
Baidu · first party $0.63 $1.25 $0.78 58 0.7s 1.05M
Together $1.74 $3.48 $2.17 53 0.8s 512k
Alibaba · first party $1.42 $2.83 $1.77 52 1.4s 1.00M
AtlasCloud $1.68 $3.38 $2.10 46 2.7s 1.05M
Wafer $1.20 $2.40 $1.50 45 5.5s 1.05M
DeepSeek · first party $0.43 $0.87 $0.54 44 1.2s 1.05M
SiliconFlow $1.50 $3.13 $1.91 41 2.1s 1.05M
StreamLake $0.67 $1.34 $0.84 41 1.6s 1.02M
CoreWeave $1.74 $3.48 $2.17 28 1.7s 1.05M
Ionstream $1.13 $2.26 $1.41 20 2.6s 1.05M
DigitalOcean $1.39 $2.78 $1.74 6 2.7s 262k
Parasail $1.74 $3.48 $2.17 — — 1.05M

Per-host pricing and telemetry from OpenRouter. Speed is output tokens per second; TTFT is time to first token.

The Known Good

History

How price and quality have moved since we began tracking this model

Quality & price over time

Since 25 Jul 2026

Known Good Index 93 · based on 3 of 6 evaluations
Blended price $0.54 per 1M · observed 25 Jul 2026
Output speed 45 tok/s · observed 25 Jul 2026
Time to first token 2.0s

No recorded changes for this model since tracking began.

The Known Good

Comparisons

Head-to-head pages pairing DeepSeek: DeepSeek V4 Pro with other tracked models

DeepSeek: DeepSeek V4 Pro vs Qwen: Qwen3.7 Max DeepSeek: DeepSeek V4 Pro vs MoonshotAI: Kimi K2.6 DeepSeek: DeepSeek V4 Pro vs Anthropic: Claude Opus 4.7 DeepSeek: DeepSeek V4 Pro vs Z.ai: GLM 5.2 DeepSeek: DeepSeek V4 Pro vs Google: Gemini 3.5 Flash DeepSeek: DeepSeek V4 Pro vs Google: Gemini 3.1 Pro Preview DeepSeek: DeepSeek V4 Pro vs Qwen: Qwen3.6 Max Preview DeepSeek: DeepSeek V4 Pro vs OpenAI: GPT-5.4 All 49 comparisons →
Sources for this page
Pricing & providers — OpenRouter Metadata — models.dev Benchmarks — Epoch AI (CC-BY 4.0)Elo — Arena Full attribution →