Hello fellow keepers of numbers,

Welcome to Super Bowl weekend! Both teams might’ve finished the regular season 14-3, but we all know the Pats are this year’s David and the Seahawks are Goliath. Which leads me into this week’s news about Anthropic (David) vs OpenAI (Goliath).

Anthropic launched an updated flagship model, Opus 4.6. OpenAI counters minutes later with GPT-5.3-Codex. Filed, the AI tax prep software, partners with Verito to bring Filed into the hosted desktop.

And lastly, I promise you’re going to want to stick around for the Claude Cowork plugins announcement and demo of a few of the new accounting skills.

THE LATEST

Anthropic announced Claude Opus 4.6, its most capable model so far, with a strong focus on coding, AI agents, and long-running enterprise workflows. Opus 4.6 introduces a 1 million token context window in beta, which lets it work over very large document sets, codebases, and datasets, and it can generate outputs up to 128,000 tokens in one response.

The model is available now in Claude’s web app, via API, and through major cloud platforms, with pricing unchanged at $5 per million input tokens and $25 per million output tokens.

Anthropic also upgraded its office integrations: Claude in Excel now handles longer, more complex spreadsheet tasks and can infer structure from messy data, while Claude in PowerPoint is launched in research preview to Max, Team, and Enterprise subscribers. Claude in PowerPoint can generate or edit slide decks that stay on-brand using existing layouts and templates.

On the developer side, Opus 4.6 adds “agent teams” in Claude Code so multiple AI agents can work in parallel on large tasks, plus new features like context compaction (automatic summarization of older context) and “effort” controls that let users trade off speed, cost, and depth of reasoning.

Why it’s important for us:

Many expected this to be the release of Claude Sonnet 5. But just when we thought Anthropic would zig, they zag. There’s a bit to unpack here:

(1) Opus 4.6: Not much to say here except the best model in the world (Opus 4.5) just got a little bit better. This update probably only makes a difference for those using Claude Code, and even then, it may be hard to notice. In my very limited testing thus far, it seems just a little slower than Opus 4.5, but better at researching and planning before completing tasks.

(2) Claude in Excel: This was already incredibly good. If you haven’t seen it in action, check out last week’s newsletter. I’ve yet to test this update, but apparently it’s even better now. That’s likely due to the Opus 4.6 upgrade.

(3) Claude in PowerPoint: This was buried in the announcement, but might be hugely beneficial for accountants. The powers of Claude in Excel are now in PowerPoint. This isn’t available to the pro plan yet, but it likely will be in the next month or two.

(4) Agent teams in Claude Code: This is probably irrelevant for most accountants, but for the Claude Code nerds, this is an interesting update. Essentially, Claude Code can now spin up several instances of Claude Code to perform tasks independently and report back to the main Claude Code session when done. It sounds like this will mostly result in difficult tasks being completed more quickly.

Overall, an amazing announcement with impressive updates by Anthropic yet again.

On the same day as Anthropic’s launch of Opus 4.6, OpenAI released GPT-5.3-Codex, a new Codex model optimized for long-running, “agentic” coding and computer-use tasks.

OpenAI says the model is about 25% faster than GPT-5.2-Codex, was used to create itself, and is designed to support workflows where the AI writes code, runs tools, and operates software over extended sessions.

GPT-5.3-Codex emphasizes interactive steering while it works and can also automate tasks like building presentations, analyzing spreadsheets, and assembling data workflows, with access rolling out through the Codex app, IDE integrations, and paid ChatGPT plans.

Why it’s important for us:

Drama. The top two AI model competitors released on the same day, within 15 minutes of each other?

Codex is essentially OpenAI’s competitor to Opus. GPT-5.2-Codex was an extremely smart model and fantastic at coding. GPT-5.3-Codex is definitely a step up. It’s much faster than its predecessor and seems to be much smarter.

The big difference between this model and Opus 4.6 for most accountants will be how OpenAI has chosen to expose it within ChatGPT. To use it, you have to click Codex in the sidebar, which launches a separate tab with a different UI. The UI strongly hints that you should use it to code (to be fair, that’s why it was built). However, Opus 4.6 sits within “regular” Claude, and you can chat with it just like any other AI model. Much more accessible than GPT-5.3-Codex.

Filed announced a partnership with Verito to make its AI-powered tax preparation workspace available directly inside Verito-managed virtual desktops. Verito customers can launch Filed from the same hosted desktop environment they already use for their tax and accounting software, with no new infrastructure or separate login flows.

Filed’s “Intelligent Tax Workspace” automates client document intake, data extraction, workpaper creation, and tax return population in existing professional tax software like Drake, UltraTax, Lacerte, ProSeries, and CCH Axcess, so preparers log into reviewer-ready returns instead of blank forms.

Verito provides private cloud hosting and managed IT built specifically for tax and accounting firms, with dedicated servers, SOC 2 controls, and compliance aligned to IRS Publication 4557 and the FTC Safeguards Rule. Filed runs inside this environment, keeping existing security, US data residency, encryption, and anonymization controls in place while adding AI automation on top of hosted tax software.

Why it’s important for us:

This is an important partnership for firms using Verito. AI and automation are sometimes harder to adopt with hosted desktops, so AI products coming to Verito customers is obviously very beneficial.

We’re probably still 1 year early on AI tax prep software, but this needs to be on firms’ radar and seriously evaluated as we head into 2027. Filed is definitely one of the stronger players in the space at the moment. Big win for Verito customers.

Anthropic announced 11 open-source plugins for Claude Cowork, turning the desktop agent into role-specific specialists for sales, legal, marketing, customer support, product management, data analysis, and finance. The plugins bundle domain knowledge, tool connections, and slash commands into packages that users can install directly from Cowork or customize for their own workflows. All 11 are available on GitHub and work with paid Claude plans.

The Finance plugin is built for month-end close and includes slash commands like /journal-entry, /reconciliation, /income-statement, /variance-analysis, and /sox-testing. It can prepare accruals, fixed asset entries, and payroll with proper debits and credits, compare GL balances to subledgers or bank statements, generate income statements with period-over-period comparison, and create SOX compliance workpapers. Users can connect ERPs, data warehouses, and spreadsheets via Model Context Protocol, or paste data and upload files directly.

Cowork runs as a research preview inside the Claude Desktop app for macOS and can read and write files in a scoped folder on your machine, executing multi-step tasks rather than just responding to chat prompts. Plugins are stored locally on the user's machine for now, with organization-wide sharing and private marketplaces coming in the following weeks. Anthropic notes that Cowork activity is not yet captured in audit logs or compliance exports and advises against using it for regulated workloads until those features ship.

Why it’s important for us:

This actually tanked the stock price of Thomson Reuters (TRI) on February 3rd. Over a 20% drop in the price of the stock, literally overnight.

The scary part is there’s nothing new here. Anthropic just decided to include some plugins for Claude Cowork that come for free with access to Cowork. Essentially, they’ve packaged up a few agent skills (Claude skills) into these plugins to give the model information on how to complete typical tasks.

I would highly recommend watching my Loom video below in the Put it to Work section. You’re going to want to see it. There’s a reason why the stock price tanked.

Agent skills are available for anyone to create, as long as you use the appropriate format. You can build upon the skills Anthropic has released in their plugins, or even adjust their existing skills to be more specific to your firm’s SOPs.

PUT IT TO WORK

I gave Claude Cowork access to a folder with several bank statements (a couple PDFs and a messy Word Doc), a credit card statement (PDF), and a few customer invoices (PDFs). I asked it to prepare a GL from scratch. After receiving the output, I asked it to create an income statement.

The result was really impressive and, honestly, a little terrifying.

Anthropic's Claude Cowork Accounting Plugin & Skills - Watch Video

WEEKLY RANDOM

Anthropic announced Claude will stay completely ad-free, taking a direct shot at OpenAI, who just weeks ago said they're bringing ads to ChatGPT for free users. Anthropic says an ad-supported model conflicts with building an assistant that actually works in your best interest, especially for sensitive topics like health, finance, and legal questions.

As part of the announcement, Anthropic early-released a few Super Bowl ads that’ll air during the game this weekend. They’re genuinely hilarious. Check one out:

You know who didn’t think they were quite as funny? Sam Altman. He had a bit of a meltdown on X in response. He admitted he laughed, but then posted an all-time rant calling Anthropic "dishonest" and "authoritarian."

If AI is going to take over the world, at least we get some fun drama before we get there.

Enjoy the Super Bowl this weekend. Let the record show my prediction: 23-19 Seahawks. Let’s see how dumb I look.

And remember, Monday may not be an “observed” holiday, but we all know offices across the country will be empty. Enjoy your 3-day weekend!