Driven Review

$20/mo for Pro gets you Claude, GPT-5 and Gemini plus a Playbook that persists strategy context across sessions. That's a real architecture decision, not a rebrand of ChatGPT with a finance skin. DrivenBench, their own leaderboard comparing models on investment tasks with cost and latency data, is the kind of receipt most AI wrappers never bother to publish.

Two things give me pause. One: it executes live trades through brokerage connectors, which is a different risk class than research summaries. Two: pricing jumps to $200/mo for 6 agents and 20x usage, and I found no free trial to de-risk that jump before committing real capital context.

Coverage is specific: US, Hong Kong, A-share equities, plus crypto, FX, options. That's a genuine differentiator against general chat tools. Whether the company sticks around to keep that data fresh is unproven.

Four tiers, $0 to $200/month, scaling on usage multiples (1x to 20x) and agent count (1 to 6). At $20/month for Pro with three portfolios per agent, this is a reasonable line item — cheaper than a data terminal seat, cheaper than an analyst-hour. The real cost control question is usage-based scaling: 5x and 20x usage tiers mean my spend depends on how aggressively agents run scheduled tasks, and I don't see caps or alerting described anywhere.

What concerns me more than price is the brokerage execution capability sitting one tier below monitoring. Portfolio monitoring and Skills-based research are audit-friendly — Playbook retains strategy context, connectors run under permission controls. But live order execution through conversational interface, on a product that explicitly disclaims personalized advice, creates a control gap I'd want closed before rollout: who approves the trade, and where's the log?

DrivenBench is the one piece that reads like real engineering discipline — publishing latency, cost, and regression data on model selection is the kind of transparency I rarely see from vendors at this price point.

Four tiers, all priced on the page. Free at $0, Pro at $20/month, Max at $100, Ultra at $200. No sales call to find the number — rare in AI finance.

TCO math: one analyst on Pro runs $240/year. Move to Max for 3 agents and 5x usage — $1,200/year. Ultra, 6 agents, 20x usage, Claude Fable 5 — $2,400/year. Team of 5 analysts on Max: $6,000/year. Storage caps (2GB-8GB) and usage multipliers mean real cost scales with actual research volume, not seat count. That's different math than most SaaS.

No published overage rate past the multiplier tiers. No free trial — only a capped free plan (1 agent, 1 portfolio, auto model only). DrivenBench gives a rare public cost/latency comparison across models, which helps justify the model-access upsell. Trades through broker connections raise the stakes on the 'no personalized advice' disclaimer — check that language before funding anything live.

The Playbook is the feature that matters here. Persistent strategy context across sessions is exactly what's missing when you're pasting the same thesis into ChatGPT every Monday morning. Pair that with Skills as structured workflows instead of free-text prompting, and scheduled tasks that auto-generate monitoring reports, and this starts to look like an actual junior analyst rather than a chatbot with a stock ticker plugin.

Day-3 reality is less clean. Free tier gives you 1 Agent, 1 portfolio, and 1GB storage — fine for kicking tires, useless for running a real book. Pro at $20/mo bumps to 3 portfolios, but you're still on one Agent, which means coverage sectors get cramped fast if you run sector rotation or multi-strategy books.

DrivenBench is a smart trust signal — publishing latency, cost, and regression disclosures against Claude, GPT-5, and Gemini beats vague "powered by AI" marketing. Trade execution through connected brokerage accounts is the real differentiator against ChatGPT-style tools, though it raises the permission-control stakes considerably.

Driven's pitch makes sense the second you read it: stop re-explaining your strategy to a chatbot every morning. The Playbook holds context, Skills structure the workflows, and scheduled tasks run monitoring without you babysitting a browser tab. That's a real answer to the 'I asked ChatGPT the same question for the fourth time' problem, and publishing DrivenBench with cost and latency numbers instead of just marketing copy is the kind of thing that builds trust slowly.

The free plan is thin though — 1GB storage, 1 agent, auto model only. You'll hit the ceiling fast and land on the $20/mo Pro tier or the $100 Max tier to get real model access. Storage caps scaling from 1GB to 8GB across tiers feels like a SaaS decision, not an investing one.

Trade execution through connected brokerages is the standout feature nobody else in this review pile has. Whether the daily experience holds up past week one, I can't say from here.

DrivenBench is a genuinely uncommon move — publishing leaderboard results, cost, latency, and regressions against your own product invites scrutiny most vendors avoid. That buys real credibility on Marketing Honesty. Claude, GPT-5, Gemini all named on the Pro tier at $20/mo, so the model claims are checkable.

Then there's live trade execution through conversation, connecting brokerage accounts and placing real orders. That's a different risk category than research summaries. I'd want a much deeper look at permission controls on Connectors before trusting that with real capital.

Exit portability is the soft spot. Playbook memory and Skills workflows are the whole value prop — and there's no stated export path for strategy context if you leave. Ultra at $200/mo for 6 agents is steep against general chat tools that do research for free, but Driven isn't really competing there. It's competing on persistent context and live order flow, which is a narrower, riskier bet.