Compare.

Raven uses the app’s shared research tools and guidance: indexed perpetuals, signals, liquidity, wallet activity, collected news, filings, macro events, saved research and evidence cards, plus web, browser, code and configured paid research. Source availability and historical gaps are preserved. Private feeds, personal watches and signed-in computer sessions are excluded from this shared demo. Codex and Claude Code labels indicate official SDK harnesses in fresh isolated workspaces, with their own prompts and public web/shell tools; personal settings and private connectors are excluded. API fallback labels are GPT and Claude. Models are shown separately, so this compares complete setups, not data access alone. Harness runs have a four-minute deadline and 24 provider requests; Claude Code also allows 12 turns. Raven allows 12 rounds and 20 custom tool calls. Costs use reported usage and tool fees, excluding infrastructure; harness times include sandbox startup. Bars show time and cost, not quality.