CoinStats AI Agent Tops Gemini, ChatGPT in Crypto Test

CoinStats’ AI Agent scored 79 out of 100 in an open-source crypto research benchmark, ahead of Google’s Gemini Deep Research (67), OpenAI’s ChatGPT Deep Research (61) and Anthropic’s Claude Deep Research (58). The company reported average response times of about four minutes for its agent versus competitors’ averages of 22 to 55 minutes. The benchmark and its evaluation criteria are published on GitHub.

The test used an AI judge to grade outputs on accuracy, depth, recency and actionability. The GitHub repository includes the benchmark methodology and scoring framework so researchers can review and reproduce the results. CoinStats provided the timing averages and score breakdowns in the project documentation.

CoinStats attributes the performance difference to direct access to crypto data. General-purpose models rely mainly on web search and publicly indexed content, while CoinStats’ system connects to onchain data, exchange-level metrics, derivatives information and real-time social sentiment feeds. The company describes those connections as sources the agent can query without routing through a general web index.

The AI Agent uses a multi-agent architecture that launches specialized workers in parallel. One worker searches real-time news, another scans social platforms, a third analyzes blockchain transactions, and others pull exchange data or check a user’s portfolio before a final worker compiles the findings. CoinStats refers to that coordination as “agentic orchestration.”

Functionally, CoinStats presents the product as a research copilot for traders and investors. The Market Research capability combines news, derivatives data, social sentiment and onchain signals to explain price moves or narrative changes. Onchain Tracking covers wallet monitoring, whale flow analysis, contract deployments and token risk scoring across more than 120 blockchains. The CoinStats Crypto API supplies onchain and portfolio data for these analyses.

Social Sentiment Analysis monitors conversations across crypto communities and influencer accounts and flags key mentions and sentiment shifts. Portfolio Analysis connects to a user’s CoinStats account to provide P&L breakdowns, suggest adjustments and answer account-specific questions such as profit on a token or the largest contributors to losses. Backtesting lets users simulate strategies against historical market data. The agent can write and run code for complex queries and produces interactive outputs such as tables and charts.

The product operates in three modes. Deep Research runs multi-step reasoning across social, onchain, technical and web sources for full reports. Backtesting executes historical strategy simulations. Fast Mode delivers quicker, lighter responses for price checks and basic queries. A separate Private Mode, implemented with Venice AI, routes queries through encrypted, decentralized infrastructure so data is not shared with third-party AI providers.

CoinStats launched the AI Agent in public beta for Degen and Premium plan users on web, iOS and Android. The company, founded by Narek Gevorgyan, said the beta release is intended to gather user feedback and guide feature development.