Comparison
Claude vs Smalt AI
for finance work.
Claude is the foundation model. Smalt AI is the finance application built on top of it. The honest framing: we use Claude (and Gemini) under the hood, then add the tooling that makes the output a real deliverable instead of a description.
The short answer
Use Claude directly (claude.ai or via API) when you want raw model access — for general reasoning, writing, coding, ad-hoc analysis, building your own workflows. Use Smalt AI when you want a finance application that produces real deliverables (Excel models, PowerPoint decks, sourced research) with the construction details handled.
Smalt AI uses Claude under the hood for most reasoning tasks — including DCF construction, research synthesis, and deck generation. We also use Gemini for fast classification (under Google's enterprise terms — no training on your data). The choice between Claude and Smalt AI is the choice between a foundation model and an application built on top of it.
Capability comparison
| Capability | Claude (claude.ai / API) | Smalt AI |
|---|---|---|
| Foundation model | Claude (Sonnet, Opus, Haiku) | Claude + Gemini (routed by task) |
| 1M-token context | Claude 1M (selected models) | Yes — used for full 10-K, multi-quarter transcripts in one pass |
| General Q&A, writing, code | Excellent | Sufficient — but not the focus |
| DCF / LBO / 3-statement Excel models | Returns markdown tables. Cells contain values. | Returns a working .xlsx with =NPV, =SUM, =XIRR formulas across 14 linked sheets |
| PowerPoint generation | Outline / markdown only | Native editable .pptx with consulting-grade typography, charts, right-rail commentary |
| Web research with citations | Possible via tool use; depends on the wrapper | Native — paragraph-level citations to filings, news, expert calls |
| Built-in finance toolkit | None | Filings parser, market data, comps lookup, deal database, financial-skills library (DCF, LBO, three-statement, sensitivity) |
| Excel file round-trip | Not native to Claude.ai | Upload an existing model, ask for changes, get the file back with formulas intact |
| Composio / OAuth tools | Not native | ~850 OAuth-gated integrations (Gmail, Sheets, Slack, HubSpot, Notion, etc.) |
| Pricing | $20/mo Pro for claude.ai; API metered by token | Free tier (500 credits, no card), Basic ~$18/mo, Pro ~$45/mo, Enterprise self-host |
| Self-hosting | API + your infra (you build the wrapper) | Enterprise self-hosting available with the application stack included |
What Smalt AI adds on top of Claude
Claude is a remarkable model — strong reasoning, excellent for long-context document analysis, the right tool for many finance tasks. But Claude alone gives you reasoning, not deliverables. The work of turning Claude's reasoning into a real Excel model, a real PowerPoint deck, or sourced research with citations sits in the layer above the model. That layer is Smalt AI.
Concretely, on top of Claude we add:
- The financial-skills library — institutional-grade workflows for DCF, LBO, three-statement, comps, returns analysis, due diligence, sensitivity, and IC memos. Each one encodes how a real analyst would build that artefact (mid-year discounting, balance-sheet checks, beginning-balance interest, 75-formula sensitivity tables, the works).
- Excel construction tooling — the ability to write a real
.xlsxwith formulas, cross-sheet references, and conditional formatting. Round-trip an existing workbook back as an actual workbook. - PowerPoint generation — the deck generator that produces consulting-grade
.pptxwith insight titles, right-rail analyst commentary, hero typography, stat-progression layouts, section navigation. Editable native PPT, not a markdown outline. - Sourced research pipeline — filings parser, news search with citations, expert call transcript ingestion, and the synthesis logic that produces an answer with paragraph-level source attribution.
- Quality checks — balance-sheet must balance, WACC must be 6–14%, terminal value 50–70% of EV, sensitivity centre cell must equal base — built into the construction process so models leave the system in a defensible state.
- Composio integration — OAuth-gated access to ~850 third-party tools (Gmail, Sheets, Slack, HubSpot, Notion, etc.) for end-to-end workflows.
- Cost discipline — credit-metered, Gemini routing for cheap classification tasks, Anthropic context editing for long sessions. Compared to building this yourself on the API, the Smalt AI tier prices reflect production-grade engineering you don't have to maintain.
When Claude direct is the right choice
Use Claude directly via claude.ai or API when:
- You're building your own workflow and want raw model access without an opinionated wrapper.
- Your work is outside the finance domain — software engineering, scientific writing, creative work, etc.
- You want to experiment with prompts and tools for your own use case.
- Your team has the engineering capacity to build the surrounding tooling (Excel construction, PPT generation, citations pipeline, etc.) — and finds that worthwhile.
When Smalt AI is the right choice
Use Smalt AI when:
- You need real deliverables — Excel files with formulas, PowerPoint decks, sourced research — not just reasoning.
- You're a finance professional (analyst, investor, banker, corp dev) and your workflows centre on models, decks, and research.
- You want the institutional-grade methodology baked in — mid-year discounting, scenario INDEX framework, sensitivity construction, balance-sheet checks — without having to encode it yourself.
- You don't want to maintain the engineering around a foundation model — you want a finished application.
Honest caveats
- Smalt AI uses Claude. We're an application layer, not a competing foundation model. If Claude has a bad day, our reasoning quality reflects it; if Anthropic ships an upgrade, we benefit.
- For pure general-purpose work, Claude direct is the cleaner choice — Smalt AI's product surface is optimised for finance, so the chrome around the chat reflects that.
- Smalt AI is younger than Claude. The model is mature; the application layer is iterating quickly.
The summary
Claude is the foundation model that does the reasoning. Smalt AI is the finance application that turns that reasoning into deliverables — real Excel models, native PowerPoint, sourced research. They're complements, not alternatives. If you're a finance professional and the deliverable matters, Smalt AI is the productivity gain. If you want raw model access for your own purposes, Claude is the right call.
500 free credits, no credit card. Try Smalt AI on a real workflow.
Built on Claude. Tuned for finance.
See what Claude plus a finance application layer can actually do.