These Standards explain how COUNCIA calculates postpaid Agent-research charges, including the current unit rates for every Agent enabled in production.
Effective and last updated: August 20, 2026 · Pricing catalog: model-pricing-2026-08-20-v10
The range shown before research is a planning estimate, not a fixed quote or price cap. The final charge is calculated only after the report completes, from the measured billable usage of each selected Agent’s final successful attempt and the successful processing operations described below.
COUNCIA currently applies the unit rates in these Standards directly, without an additional percentage markup. Provider list prices, product SKUs, models, and billing units differ, so two Agents with similar visible answers may have different final charges.
For a completed report, all verified calls belonging to each Agent’s final successful attempt are added together. Failed attempts, retry work, failed Agents, failed runs, and runs cancelled before an Agent starts are not charged. If COUNCIA cannot verify a final amount under the applicable catalog and usage record, that usage is waived rather than estimated.
Only provider-billable model Tokens are used for Token-priced Agents. Search, connector, or tool activity is charged separately only where the table states a per-call or per-request rate. A report’s final total is the sum of Agent charges and eligible processing charges, not the sum of the pre-run estimates.
The following 15 Agents are enabled in production on the effective date above. “Per 1M Tokens” means per one million provider-billable Tokens in the stated category.
| Agent / model | Current unit rate | Additional rules | Provider rate source |
|---|---|---|---|
| Baidu AI SearchBaidu AI Search High Performance | CNY 0.06 / request | Fixed request price; Token volume does not add another charge. | Provider rate source ↗ |
| Claude CodeClaude Opus 5 | US$5 input / $0.50 cache read / $6.25 5-minute cache write / $10 1-hour cache write / $25 output per 1M Tokens | The primary model and the model currently observed in production billing are both Claude Opus 5. Web search is US$0.01/request; only measured usage from the final successful attempt is charged. | Provider rate source ↗ |
| CodexGPT-5.6 Sol | US$5 input / $0.50 cache read / $6.25 cache write / $30 output per 1M Tokens | When a request exceeds 272K input Tokens: US$10 input / $1 cache read / $12.50 cache write / $45 output per 1M. Web search is US$0.01/call. | Provider rate source ↗ |
| DeepSeek HarnessDeepSeek V4 Pro | US$0.435 input / $0.003625 cache hit / $0.87 output per 1M Tokens | Only usage from the final successful attempt is included. | Provider rate source ↗ |
| Doubao Assistant APIDoubao Assistant reasoning_search | CNY 0.50 / call | This is the fixed postpaid SKU shown by the production Volcengine account; a zero Token count does not make the successful call free. | Provider rate source ↗ |
| Volcengine Online Q&AOnline Q&A Agent Generic Lite | CNY 9 input / CNY 22 output per 1M Tokens | Uses the standalone Online Q&A Agent SKU, not a Doubao Ark model proxy price. | Provider rate source ↗ |
| Gemini 3.7 Flash ResearchGemini 3.7 Flash | US$0.75 input / $0.075 cache read / $3.75 output per 1M Tokens | Separately reported reasoning Tokens are added to billable output. Grounded search is US$0.014/query. These are the current introductory Standard rates through December 31, 2026. | Provider rate source ↗ |
| Grok 4.6 ResearchGrok 4.6 | US$2 input / $0.50 cache read / $6 output per 1M Tokens | At 200K or more input Tokens: US$4 input / $1 cache read / $12 output per 1M. Web search, X search, and code execution are US$0.005/call. | Provider rate source ↗ |
| Kimi CodeKimi K3 | US$3 input / $0.30 cache read / $15 output per 1M Tokens | Only usage from the final successful attempt is included. | Provider rate source ↗ |
| Mistral ResearchMistral Large | US$0.50 input / $0.05 cache read / $1.50 output per 1M Tokens | Web search and code-interpreter tools are US$0.03/call. | Provider rate source ↗ |
| Perplexity ResearchPerplexity Agent API · Medium | The exact USD amount returned by the Perplexity Agent API | COUNCIA does not replace this with a Token estimate. If a verifiable amount is not returned, the Agent charge is waived. | Provider rate source ↗ |
| Perplexity FinancePerplexity Sonar Finance Search | The exact USD amount returned by the Perplexity Agent API | The provider amount may reflect model and finance-search usage. If it cannot be verified, the Agent charge is waived. | Provider rate source ↗ |
| Qwen Web SearchQwen Web Search Agent | CNY 0.022 / request | Fixed request price; Token volume does not add another charge. | Provider rate source ↗ |
| WorkBuddy Investment ResearchTencent Hy3 public API reference | US$0.132 input / $0.033 cache read / $0.528 output per 1M Tokens | Uses the public Hy3 API replacement rate; promotional WorkBuddy Credit multipliers do not affect user billing. | Provider rate source ↗ |
| Zhipu Web SearchGLM-5.3 | CNY 8 input / CNY 2 cache read / CNY 28 output per 1M Tokens | The current Search Pro engine is CNY 0.03/request; Standard is CNY 0.01, and Sogou or Quark Pro is CNY 0.05 when actually reported. | Provider rate source ↗ |
The first research run also includes successful model usage already incurred to structure, refine, review, and translate that question. Each successful Agent answer may require a separate bilingual translation pass. These operations currently use DeepSeek V4 Pro at US$0.435 input, US$0.003625 cache hit, and US$0.87 output per one million Tokens.
Question-preparation usage is attached only to the first run created from that question draft. A later rerun does not charge those original preparation calls again, although new successful translation or other processing calls created for that rerun may be charged. Unverifiable processing usage is waived.
The COUNCIA wallet and final research charge are denominated in U.S. dollars. CNY-denominated usage is converted using COUNCIA’s accepted USD/CNY billing quote for that usage record. The native amount, exchange rate, source, timestamp, and pricing versions are retained so the charge can be audited later.
Individual usage records are kept at high precision. COUNCIA adds verified Agent and processing amounts in U.S. dollars, then rounds the completed run total once to the nearest U.S. cent using decimal round-half-up. There is no per-Agent one-cent minimum final charge; the one-cent minimum shown in some estimates is only a display convention.
Planning ranges use recent comparable successful runs under the current catalog when enough data exists; otherwise they use a fallback range. Actual usage can be below or above that range and may make the account balance negative. New research is blocked when the balance is below the displayed minimum.
COUNCIA may update Agent rosters, models, provider rates, supported tools, and pricing methods prospectively. Usage records retain the catalog and exchange-rate versions applied to them. Refunds, duplicate charges, inaccessible services, and non-waivable consumer rights are governed by the Refund Policy and Terms of Service.
