DeepSeek: DeepSeek V3.1 provider comparison

deepseek/deepseek-chat-v3.1

DeepSeek-V3.1 is a large hybrid reasoning model (671B parameters, 37B active) that supports both thinking and non-thinking modes via prompt templates. It extends the DeepSeek-V3 base with a two-phase long-context...

Endpoints:: 8
Context:: 163.8K
Input:: text
Output:: text

Compare this model in a set →

At a glance

Comparison winners

Based on 1,000 input and 500 output tokens using the latest published median speed data.

Best cost/speed trade-off

NovitaAInovita/fp8

$0.5133 / 1M tokens

25.54 s estimated response

Cheapest

DeepInfradeepinfra/fp4

$0.4833 / 1M tokens

$0.000725 for the sample request

Fastest

Google Vertexgoogle-vertex/us-west2

5.96 s estimated response

$0.9667 / 1M tokens

Visual comparison

Price and performance charts

Compare published endpoint pricing and estimated response time visually.

Effective price by endpoint

USD per 1 million tokens for input and output. Lower is better.

DeepInfra has the lowest estimated cost for 1,000 input and 500 output tokens at $0.000725.

Estimated cost vs. response time

Based on 1,000 input and 500 output tokens. Cost is shown in USD per 1 million tokens. Lower and further left is better; the single best cost/speed trade-off appears at full opacity.

8 providers have complete pricing and speed data. NovitaAI is closest to the ideal combination of lowest cost and fastest response.

Server-rendered comparison

Provider endpoints

Prompt and completion prices are effective USD per token as published by OpenRouter. Missing speed data never removes an endpoint.

Rank providers by

Endpoint comparison for DeepSeek: DeepSeek V3.1
Provider / route	Current pricing ↗	Speed measurements	Endpoint details
Provider / route DeepInfra`deepinfra/fp4` Cheapest	Current pricing Input price $0.25 / 1M tokens Output price $0.95 / 1M tokens Sample cost $0.000725	Speed measurements 500-token response 56.64 s Response starts 1.08 s Output speed 9.0 tok/s	Endpoint details Uptime (30 min) 99.95% Context 163.8K Model format fp4 Max output 32.8K 17 parameters frequency_penalty, include_reasoning, logit_bias, max_tokens, min_p, presence_penalty, reasoning, repetition_penalty, response_format, seed, stop, structured_outputs, temperature, tool_choice, tools, top_k, top_p
Provider / route NovitaAI`novita/fp8` Best cost/speed trade-off	Current pricing Input price $0.27 / 1M tokens Output price $1 / 1M tokens Sample cost $0.00077	Speed measurements 500-token response 25.54 s Response starts 1.73 s Output speed 21.0 tok/s	Endpoint details Uptime (30 min) 99.65% Context 131.1K Model format fp8 Max output 32.8K 15 parameters frequency_penalty, include_reasoning, max_tokens, presence_penalty, reasoning, repetition_penalty, response_format, seed, stop, structured_outputs, temperature, tool_choice, tools, top_k, top_p
Provider / route SiliconFlow`siliconflow/fp8`	Current pricing Input price $0.27 / 1M tokens Output price $1 / 1M tokens Sample cost $0.00077	Speed measurements 500-token response 29.41 s Response starts 1.63 s Output speed 18.0 tok/s	Endpoint details Uptime (30 min) 97.38% Context 163.8K Model format fp8 Max output 163.8K 11 parameters frequency_penalty, include_reasoning, max_tokens, reasoning, response_format, structured_outputs, temperature, tool_choice, tools, top_k, top_p
Provider / route AtlasCloud`atlas-cloud/fp8`	Current pricing Input price $0.3 / 1M tokens Output price $0.95 / 1M tokens Sample cost $0.000775	Speed measurements 500-token response 29.02 s Response starts 1.25 s Output speed 18.0 tok/s	Endpoint details Uptime (30 min) 99.16% Context 131.1K Model format fp8 Max output 65.5K 15 parameters frequency_penalty, include_reasoning, logit_bias, max_tokens, min_p, presence_penalty, reasoning, repetition_penalty, response_format, seed, stop, structured_outputs, temperature, top_k, top_p
Provider / route Weights & Biases`wandb/fp8`	Current pricing Input price $0.55 / 1M tokens Output price $1.65 / 1M tokens Sample cost $0.001375	Speed measurements 500-token response 10.78 s Response starts 0.36 s Output speed 48.0 tok/s	Endpoint details Uptime (30 min) 100.00% Context 161K Model format fp8 Max output 161K 17 parameters frequency_penalty, include_reasoning, logprobs, max_tokens, presence_penalty, reasoning, repetition_penalty, response_format, seed, stop, structured_outputs, temperature, tool_choice, tools, top_k, top_logprobs, top_p
Provider / route Google Vertex`google-vertex/us-west2` Fastest	Current pricing Input price $0.6 / 1M tokens Output price $1.7 / 1M tokens Sample cost $0.00145	Speed measurements 500-token response 5.96 s Response starts 0.86 s Output speed 98.0 tok/s	Endpoint details Uptime (30 min) 100.00% Context 163.8K Model format unknown Max output 32.8K 17 parameters frequency_penalty, include_reasoning, logprobs, max_tokens, presence_penalty, reasoning, repetition_penalty, response_format, seed, stop, structured_outputs, temperature, tool_choice, tools, top_k, top_logprobs, top_p
Provider / route MARA`mara`	Current pricing Input price $0.6 / 1M tokens Output price $1.7 / 1M tokens Sample cost $0.00145	Speed measurements 500-token response 21.10 s Response starts 2.58 s Output speed 27.0 tok/s	Endpoint details Uptime (30 min) 55.44% Context 131.1K Model format unknown Max output 7.2K 11 parameters include_reasoning, logprobs, max_tokens, reasoning, response_format, stop, structured_outputs, temperature, top_k, top_logprobs, top_p
Provider / route SambaNova`sambanova/fp8`	Current pricing Input price $0.65 / 1M tokens Output price $1.5 / 1M tokens Sample cost $0.0014	Speed measurements 500-token response 10.60 s Response starts 0.98 s Output speed 52.0 tok/s	Endpoint details Uptime (30 min) 99.64% Context 131.1K Model format fp8 Max output 7.2K 9 parameters include_reasoning, max_tokens, reasoning, stop, temperature, tool_choice, tools, top_k, top_p

Endpoint data fetched Jul 15, 2026, 11:17 PM UTC.

DeepSeek V3.1 deployment guide

How to choose a DeepSeek V3.1 inference provider

DeepSeek V3.1 is indexed here as a text model by DeepSeek, with 8 provider endpoints from DeepInfra, NovitaAI, SiliconFlow, AtlasCloud, Weights & Biases, and others. The comparison preserves each exact OpenRouter routing tag so pricing and performance observations can be connected to the route an application would actually request.

Match the endpoint to the workload

The model publishes a 163.8K-token context window. It accepts text input and returns text output. 19 distinct supported parameters appear across the listed routes. Confirm limits on the specific endpoint rather than assuming every host exposes the same configuration.

Compare the real request economics

DeepInfra currently has the lowest estimated cost for the standard 1,000-input/500-output-token sample at $0.000725. Input-heavy and output-heavy applications can produce a different result, so review both per-million-token prices in the endpoint table.

Balance response start and generation speed

Google Vertex currently has the shortest estimated 500-token response at 5.96 seconds. NovitaAI is the single endpoint closest to the current ideal cost/speed combination. Recent observations can change, so validate finalists with your own prompts.