Models

Models

Every model Laras Router serves through its OpenAI-compatible endpoint, with the limits the API enforces and the price it charges. Your tools can read the same figures from GET /v1/models.

ModelContextMax outputInput / 1MOutput / 1M
DeepSeek-V4.1-Flash1M384K$0.400$1.600
DeepSeek-V4-Pro1M384K$1.760$5.280
GLM-5.3-Flash1M128K$0.200$0.667

by DeepSeek

DeepSeek-V4.1-Flash

Get an API key
Context window
1Mtokens
Prompt and output together
Maximum output
384Ktokens
max_tokens up to 384,000
Default output cap
8,192tokens
When max_tokens is omitted

Pricing

USD per 1M tokens · the same price every hour

Input, cache hitPrompt tokens the provider already has in its cache
$0.010/ 1M
Input, cache missPrompt tokens processed fresh
$0.400/ 1M
OutputGenerated tokens, reasoning included
$1.600/ 1M

How prompt caching works

Caching is automatic. When a request starts with the same tokens as a recent one, such as a system prompt, tool definitions, or earlier turns of a conversation, those tokens are billed at the cache-hit rate. Each response reports the split in prompt_cache_hit_tokens and prompt_cache_miss_tokens.

Estimate your costHow billing works

Capabilities

  • Streaming Server-Sent Events with stream: true
  • Tool calling tools, including parallel calls
  • JSON output response_format set to JSON object
  • Reasoning Returned in reasoning_content, billed as output
  • Image input

These are the values GET /v1/models returns for deepseek-flash, price version deepseek-flash-2026-09-v1.

Model guideAPI reference

by DeepSeek

DeepSeek-V4-Pro

Model version DeepSeek-V4-Pro-0813

Get an API key
Context window
1Mtokens
Prompt and output together
Maximum output
384Ktokens
max_tokens up to 384,000
Default output cap
8,192tokens
When max_tokens is omitted

Pricing

USD per 1M tokens · the same price every hour

Input, cache hitPrompt tokens the provider already has in its cache
$0.060/ 1M
Input, cache missPrompt tokens processed fresh
$1.760/ 1M
OutputGenerated tokens, reasoning included
$5.280/ 1M

How prompt caching works

Caching is automatic. When a request starts with the same tokens as a recent one, such as a system prompt, tool definitions, or earlier turns of a conversation, those tokens are billed at the cache-hit rate. Each response reports the split in prompt_cache_hit_tokens and prompt_cache_miss_tokens.

Estimate your costHow billing works

Capabilities

  • Streaming Server-Sent Events with stream: true
  • Tool calling tools, including parallel calls
  • JSON output response_format set to JSON object
  • Reasoning Returned in reasoning_content, billed as output
  • Image input A message with an image returns 400 unsupported_content

These are the values GET /v1/models returns for deepseek-v4-pro, price version deepseek-v4-pro-2026-10-v1.

Model guideAPI reference

by Z.ai

GLM-5.3-Flash

Model version GLM-5.3-Flash-2026-11-08

Get an API key
Context window
1Mtokens
Prompt and output together
Maximum output
128Ktokens
max_tokens up to 128,000
Default output cap
8,192tokens
When max_tokens is omitted

Pricing

USD per 1M tokens · the same price every hour

Input, cache hitPrompt tokens the provider already has in its cache
$0.040/ 1M
Input, cache missPrompt tokens processed fresh
$0.200/ 1M
OutputGenerated tokens, reasoning included
$0.667/ 1M

How prompt caching works

Caching is automatic. When a request starts with the same tokens as a recent one, such as a system prompt, tool definitions, or earlier turns of a conversation, those tokens are billed at the cache-hit rate. Each response reports the split in prompt_cache_hit_tokens and prompt_cache_miss_tokens.

Estimate your costHow billing works

Capabilities

  • Streaming Server-Sent Events with stream: true
  • Tool calling tools, including parallel calls
  • JSON output response_format set to JSON object
  • Reasoning Returned in reasoning_content, billed as output
  • Image input
Per-key limits Each API key starts at 300 requests per minute and 16 concurrent requests, for every model.
Provider updates: Model availability and pricing may change as providers release new versions or revise their service. Prices are versioned, and GET /v1/models always returns the version in effect. Legacy model IDs such as deepseek-v4-flash are not accepted and return 404 model_not_found; use deepseek-flash.