Skip to content
RunicGet Runic

Runic Docs Reference

Qwen provider

⌘ K

Search stays on this site. No external service.

Qwen (DashScope / Alibaba Cloud) is API-token based. No browser cookies. Fetches per-model token consumption with estimated USD cost.

Data sources + fallback order

  1. API token (single source, one fetch strategy).

Credentials

Token resolution order:

  1. Keychain token (stored from Preferences).
  2. Environment variable DASHSCOPE_API_KEY.

The env var value is cleaned by QwenSettingsReader: whitespace trimmed, surrounding quotes stripped.

Keychain location

  • Service: com.sriinnu.athena.Runic.provider-credentials.v2
  • Account: qwen-api-token

API endpoints

  • GET https://dashscope.aliyuncs.com/api/v1/usage
  • Headers:
    • Authorization: Bearer <token>
    • Accept: application/json
    • Content-Type: application/json
  • Timeout: 15 seconds.

Parsing + mapping

Response structure

Top-level response (QwenUsageResponse):

  • code – success check (nil, "200", or "Success").
  • message – error message on failure.
  • output – main usage payload (preferred).
  • usage – alternative usage payload (fallback).

Per-model entries from output.models[] (or usage.models[]):

  • model_id / model – model identifier.
  • input_tokens, output_tokens – per-direction token counts.
  • total_tokens / tokens – total tokens (computed as input+output if both nil).
  • request_count / requests / calls – request count.

Totals

API-reported totals (preferred over summing model entries):

  • output.total_input_tokens / usage.input_tokens – total input tokens.
  • output.total_output_tokens / usage.output_tokens – total output tokens.
  • output.total_tokens / usage.total_tokens – total tokens.

Cost estimation

Static pricing table (QwenModelPricing) covers 18 models including qwen-turbo, qwen-plus, qwen-max, qwen-vl-, qwen-long, qwen2.5- (7b/14b/32b/72b), and qwen2.5-coder-*. Pricing lookup is case-insensitive with fuzzy substring matching as a fallback.

Cost formula: (input_tokens / 1M) * input_price + (output_tokens / 1M) * output_price. When only total tokens are available, a 60/40 input/output split is assumed.

Usage snapshot

  • usedPercent: 0, hasKnownLimit: false – DashScope reports consumption with no hard cap, so there is no denominator for a percentage.
  • Summary text: formatted token count + request count + estimated USD cost (e.g., “12.5K tokens · 47 req · ~$0.03 est.”).
  • Model entries sorted by total tokens descending for per-model breakdown.

Key files

  • Sources/RunicCore/Providers/Qwen/QwenProviderDescriptor.swift
  • Sources/RunicCore/Providers/Qwen/QwenSettingsReader.swift
  • Sources/RunicCore/Providers/Qwen/QwenUsageStats.swift

Read the source ↗ Checked against Runic 2.13.5 · 9730f91