algofin 1.0

The future of financial reasoning.

A model that answers numerical questions about company annual reports. Financial firms license it and run it on their own servers.

Talk to us about algofin
  • 90.9%on ConvFinQA, ahead of GPT-5.5 and Claude Opus 4.7
  • 10xcheaper per task than GPT-5.5
  • 0bytes of your data sent to OpenAI, Anthropic or Google

Use cases

Anywhere your team works out numbers from company reports. algofin 1.0 quotes the figures it uses and shows the calculation.

  • Equity research

    Growth rates, margins and year-on-year changes, taken from a company's reported tables and calculated in code.

    what was the percent change in revenue from 2007 to 2008?

  • Credit analysis

    Leverage, interest cover and liquidity ratios, with each input quoted from the statement it came from.

    what is long-term debt as a share of total capital?

  • Due diligence

    Work through a target's annual reports figure by figure, without the documents leaving your network.

    what share of revenue came from the largest segment?

  • Audit and assurance

    Recompute the numbers behind a reported figure, with the working shown so a reviewer can follow each step.

    what was the net change in the allowance for doubtful accounts?

  • Follow-up questions

    Ask a question, then ask about the answer. It keeps track of what “they” and “the change” refer to.

    and what was it the year before?

  • Regulated teams

    AI on filings and client documents for teams that cannot send them to an outside provider. It runs air-gapped if it has to.

    how much of the portfolio sits in level 3 assets?

ConvFinQA leaderboard

algofin 1.0 scores 90.9%, ahead of GPT-5.5 and Claude Opus 4.7.

The low end of our 95% interval, 89.0, is still above GPT-5.5.

algofin 1.0
90.9
GPT-5.5
88.0
Claude Opus 4.7
86.0
DeepSeek V4 Pro
83.0
Gemini 3.1 Pro
82.0
Grok 4.3
79.0
Kimi K2.6
78.0
GLM-5.1
75.0
MiniMax M2.7
73.0
ConvFinQA accuracy, %. Other models: FinanceBenchmark.ai leaderboard, retrieved 30 Sep 2026.

Cost per task

Outscores GPT-5.5 for a tenth of the price.

10xcheaper per task

GPT-5.5$0.016
algofin 1.0$0.0016

GPT-5.5 at OpenAI list price. algofin 1.0 self-hosted on one GPU.

Self-hosted

algofin 1.0 runs on your own servers. Your filings, prompts and answers never leave your network.

Your network

InYour filings and reports

algofin 1.0

on your own hardware

OutAnswers, with the calculation shown

Never sent to

  • OpenAI
  • Anthropic
  • Google
  • Anyone else

Licensing

Financial firms license algofin 1.0 and run it on their own infrastructure.

Hosted AI APIalgofin 1.0
Runs onThe provider's serversYour servers
Your dataSent to the providerStays on your network
Model changesOn their scheduleWhen you choose
Runs air-gappedNoYes

Run algofin 1.0 on your own servers.

Book a call and we will walk you through licensing it for your firm.

As seen in