DeepSeek V4

deepseek/deepseek-v4

by DeepSeek·Text·164K context·Released 2026-02-18·MIT

open-weightreasoning

Overview

DeepSeek's full V4 flagship, a large MoE with strong reasoning, coding and math, released under a permissive license. It competes with first-party frontier models on hard benchmarks at a fraction of the price, which has made it a favorite for cost-sensitive teams running complex work. Multiple providers host it, so LLMRouteHub can route for price and uptime. Heavier and slower than the Flash distill, but markedly more capable.

Benchmarks

Standardized scores (0–100), normalized across the catalog.

Intelligence85

Coding84

Instruction following90

Providers

LLMRouteHub routes to the cheapest healthy provider by default. Showing the top 5 of 8 providers by routed volume.

Provider	Input	Output	Throughput	Uptime
Groq	$0.58	$2.30	64 tok/s	99.5%
Lambda	$0.70	$2.80	60 tok/s	99.4%
Hyperbolic	$0.75	$2.98	55 tok/s	99.2%
Parasail	$0.88	$3.48	51 tok/s	99.1%
DeepInfra	$1.00	$3.99	46 tok/s	98.9%

Use this model

curl https://llmroutehub.com/api/v1/chat/completions \
  -H "Authorization: Bearer $LLMROUTEHUB_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek/deepseek-v4",
    "messages": [{ "role": "user", "content": "Hello" }]
  }'