Question 1

What is DeepSeek V3.1?

Accepted Answer

DeepSeek V3.1 is a model from Deepseek, released on Aug 21, 2025. Additionally, DeepSeek-V3.1 is trained using the **UE8M0 FP8 scale data format on both model weights and activations** to ensure compatibility with microscaling data formats. Please refer to DeepGEMM for more details.

Question 2

How much does DeepSeek V3.1 cost?

Accepted Answer

On the Respan gateway, input starts at $0.25 per 1M tokens on DeepInfra and OpenRouter, and output starts at $0.95 per 1M tokens on DeepInfra and OpenRouter, after gateway discounts. Averaged across 6 providers, it costs $0.43 per 1M input tokens and $1.33 per 1M output tokens.

Question 3

Which providers serve DeepSeek V3.1 on Respan?

Accepted Answer

6 providers serve DeepSeek V3.1: DeepInfra, OpenRouter, Featherless, Tera, TogetherAI, and Google Vertex AI. Tera, TogetherAI, and Google Vertex AI require your own provider API key (BYOK).

Question 4

What is DeepSeek V3.1's context window?

Accepted Answer

DeepSeek V3.1 accepts up to 164K tokens of context on its largest provider route.

Question 5

How do I use DeepSeek V3.1 with the Respan gateway?

Accepted Answer

Point any OpenAI-compatible SDK at https://api.respan.ai/api/ with your Respan API key and set the model to "deepseek-v3.1". To set it up with a coding agent, copy the agent setup prompt at the top of this page.

Question 6

How fast is DeepSeek V3.1?

Accepted Answer

Over the last 30 days of requests through the Respan gateway, the fastest provider was OpenRouter at 16 tokens per second, and the lowest response latency was 2.4 seconds on OpenRouter. Speed varies with request size and provider.

Question 7

Where does DeepSeek V3.1 rank on the Respan Index?

Accepted Answer

DeepSeek V3.1 scores 16.7 out of 100 on the Respan Index v1.1, ranking #115 of 138 models. By category: Coding 5.2, Agents 12.3, Reasoning 15.1, Knowledge & language 15.5 (estimated), and Human preference 46.6. The Respan Index combines independent and lab-reported benchmark results into one score, using benchmark data collected by llmmetric.com.

Question 8

How does DeepSeek V3.1 score on benchmarks?

Accepted Answer

DeepSeek V3.1 has 18 published benchmark scores, including Aider-Polyglot (thinking) 76.3%, SimpleQA (search agent) 93.4%, and GPQA-Diamond (thinking) 80.1%. Each score links to where it was published; the data is collected by llmmetric.com.

Provider	Context	Input	Output
DeepInfra`deepinfra/deepseek-ai/DeepSeek-V3.1`	164K	$0.25/M	$0.95/M
OpenRouter`openrouter/deepseek/deepseek-chat-v3.1`	164K	$0.25/M	$0.95/M
Featherless`featherless_ai/deepseek-ai/DeepSeek-V3.1`	131K	$0.28/M	$1.00/M
TeraBYOK only`tera/deepseek-ai/DeepSeek-V3.1`	164K	$0.60/M	$1.70/M
TogetherAIBYOK only`together_ai/deepseek-ai/DeepSeek-V3.1`	131K	$0.60/M	$1.70/M
Google Vertex AIBYOK only`vertex_ai/deepseek-ai/deepseek-v3.1-maas`	164K	$0.60/M	$1.70/M

DeepSeek V3.1

Providers

Certified where it counts.

Pricing

Cost by volume

Throughput and latency

Output speed

Response latency

Benchmarks

Respan Indexv1.1

Coding4

Apps

DeepSeek V3.1 FAQ

Works with your entire stack

Built for AI agents.
Break less.
Ship more.

Agents4

Reasoning4

Math3

Knowledge2

Arena1