Models

Laras Router serves two models: deepseek-flash (DeepSeek-V4.1-Flash) and deepseek-v4-pro (DeepSeek-V4-Pro). Send the one you want as "model": "deepseek-flash" or "model": "deepseek-v4-pro".

deepseek-flash

deepseek-v4-pro

Pricing

Prices are per 1M tokens, in USD, and do not change by time of day.

The provider caches repeated prompt prefixes automatically. The usage object of each response tells you how many input tokens were a cache hit (prompt_cache_hit_tokens) and how many were not (prompt_cache_miss_tokens). See Billing for a worked example.

Output length

If you omit max_tokens, the gateway applies a cap of 8,192 output tokens. Reasoning tokens count toward the cap, so a long answer to a hard question can stop with finish_reason: "length". Raise max_tokens when you expect long output — up to 384,000. A larger value is rejected with 400 max_tokens_exceeded; it is never reduced silently.

max_completion_tokens is accepted as an alias. If you send both, the smaller value applies.

What is not supported

• Images and other non-text input. A message with an image part is rejected with 400 unsupported_content.
• Other model IDs. Names such as deepseek-v4-flash or gpt-4o return 404 model_not_found. Use deepseek-flash or deepseek-v4-pro.

List models from the API

GET /v1/models returns the models your key can call, in the OpenAI list format:

curl https://api.larasrouter.com/v1/models \
  -H "Authorization: Bearer $LARAS_API_KEY"

{
  "object": "list",
  "data": [
    {
      "id": "deepseek-flash",
      "object": "model",
      "created": 1760241600,
      "owned_by": "larasrouter",
      "context_length": 1000000,
      "max_output_tokens": 384000,
      "default_output_tokens": 8192,
      "capabilities": { "streaming": true, "tools": true, "json_output": true, "thinking": true, "vision": false },
      "pricing": { "version": "deepseek-flash-2026-09-v1", "currency": "USD", "unit_tokens": 1000000, "input_cache_hit": "0.010", "input": "0.400", "output": "1.600" }
    }
  ]
}

Each entry also carries the model's limits (context_length, max_output_tokens, default_output_tokens), its capabilities, and its pricing in USD per 1M tokens, so a tool can configure itself from the API. See the List models reference for the full object.