Quick answer:
"Claude for Finance" is not one single product Anthropic launched on one date. It is the name Anthropic's own marketing now uses for a stack it has built up in stages since July 2025: pre-built Agent Skills inside a Claude for Excel research preview, a family of Claude Cowork plugins, and ten ready-to-run agent templates for financial-services work. A widely watched 19 August 2026 walkthrough video calls this "8 skills," but Anthropic's own open-source finance plugin, the one that video actually installs and demonstrates, documents 6 skills: journal entry preparation, reconciliation, financial statements, variance analysis, close management and audit support. A separate set of 6 Agent Skills, for comps, DCF models, due diligence packs, teasers, earnings analyses and initiating-coverage reports, ships inside Claude for Excel. Twelve real, named skills exist across Anthropic's own sources; no Anthropic source confirms a figure of 8.
Finance teams are exactly the kind of repetitive, document-heavy, rules-bound environment that Anthropic's Agent Skills were built for, and Anthropic has spent over a year building specifically for that audience. The trouble is that the coverage has moved faster than the product line stayed still: each new release stacks on top of, rather than replaces, the last one, and headline figures from different months describe entirely different skill sets.
This article works from Anthropic's own newsroom, its solutions pages, its open-source GitHub repository, and its official Agent Skills documentation, cross-checked against independent reporting and hands-on YouTube coverage, to lay out exactly what exists, what it costs in the UK, how it technically works, and precisely where the "8 skills" framing does and doesn't hold up.
Hands-on coverage of the release from 19 August 2026, installing and walking through Anthropic's open-source finance plugin, including its reconciliation, variance-analysis and month-end-close skills.
Executive summary
- Claude for Finance is a marketing umbrella, not a single launch. It spans a 15 July 2025 initial release, a 27 October 2025 expansion, a 24 February 2026 Cowork-plugins update and a 5 May 2026 agent-templates release, all still active on Anthropic's solutions pages as of 20 August 2026.
- Two separate, verifiable Agent Skills sets exist, six skills each. One set (journal-entry-prep, reconciliation, financial-statements, variance-analysis, close-management, audit-support) is a Claude Cowork/Claude Code plugin for accounting operations, open-sourced on GitHub. The other (comparable company analysis, DCF models, due diligence data packs, company teasers, earnings analyses, initiating coverage reports) is built into the Claude for Excel research preview.
- The "8 skills" figure traces to a single YouTube video, published 19 August 2026, that demonstrates the 6-skill Cowork plugin but titles itself "8 Skills." We could not verify a figure of 8 against any Anthropic source.
- Claude Opus 4.7 leads Anthropic's cited benchmark at 64.37% on Vals AI's independent Finance Agent benchmark, up from Sonnet 4.5's 55.3% cited seven months earlier, a jump journalists have flagged as still leaving considerable room for error in unsupervised use.
- There is no separate Claude for Finance price tag. Everything runs on Anthropic's standard Pro, Max, Team and Enterprise plans, plus, in some cases, separately licensed data-provider connectors.
- Named customers span the full spectrum of finance, from Citadel, BNY and Citi to Bridgewater, Commonwealth Bank of Australia and AIG, though Anthropic has not published adoption numbers or measured productivity gains for any of them.
What Claude for Finance actually is
Start with the timeline, because it explains almost everything about why coverage of this product is so inconsistent. Anthropic first announced Claude for Financial Services on 15 July 2025: a unified interface for querying market data and internal platforms such as Databricks and Snowflake through Claude 4 models, with MCP connectors to data providers and named early customers including Bridgewater's AIA Labs, Commonwealth Bank of Australia and AIG.
On 27 October 2025, Anthropic published Advancing Claude for Financial Services, adding a Claude for Excel research preview and a first set of 6 pre-built Agent Skills aimed squarely at investment-banking and equity-research workflows. That article cited Claude Sonnet 4.5 topping Vals AI's Finance Agent benchmark at 55.3% accuracy, and named a much longer customer list: Citi, RBC Capital Markets, Brex, Block, Coinbase, British Columbia Investment Management Corporation, Visa, Jump Trading, Francisco Partners, Chronograph, Moody's and the London Stock Exchange Group.

On 24 February 2026, a separate Cowork and plugins for finance post described five different Anthropic-built Cowork plugins, Financial Analysis, Investment Banking, Equity Research, Private Equity and Wealth Management, plus partner plugins from LSEG and S&P Global, available across Claude's paid plans on Mac and Windows.
Then on 5 May 2026, Anthropic published Agents for financial services, its most substantial update: ten ready-to-run agent templates, shipped as plugins in Claude Cowork and Claude Code, as Claude Managed Agents cookbooks, and as copyable reference code. That release cited Claude Opus 4.7 reaching 64.37% on the same Vals AI benchmark and named a fresh set of institutional customers: Citadel, FIS, BNY, Carlyle, Mizuho, Travelers, Walleye Capital and Hg, alongside a long list of data-provider partners including FactSet, S&P Capital IQ, MSCI, PitchBook, Morningstar and Moody's.
"Claude for Finance," as a phrase, is what Anthropic's own claude.com/solutions/financial-services page now uses to describe the combined result of all four releases. It is a real, current, actively maintained product line, spanning banking, insurance and asset management. It is just not a single thing that was "dropped" on 19 August 2026, as some recent coverage frames it; that date is when a YouTube creator published a detailed walkthrough of one specific piece of it, the open-source Cowork finance plugin covered in the next section, roughly three and a half months after Anthropic itself shipped it.
The skills, explained
Two genuinely distinct sets of Agent Skills exist inside the Claude for Finance umbrella, built for different jobs and different people. Neither, on its own, adds up to 8.
Set one: the six accounting-operations skills (Claude Cowork / Claude Code)
This is the open-source finance plugin published in Anthropic's public anthropics/knowledge-work-plugins GitHub repository, installable with claude plugins add knowledge-work-plugins/finance. It targets corporate accounting and Office-of-the-CFO work rather than deal-making, and it is the plugin the 19 August 2026 video actually installs and demonstrates on screen. Its README.md lists exactly 6 skills:
| Skill | What it covers |
|---|---|
| journal-entry-prep | Journal entry best practices: standard accrual types, supporting documentation requirements, and review workflows. |
| reconciliation | Methodology for GL-to-subledger, bank, and intercompany reconciliations, including reconciling-item categorisation and ageing. |
| financial-statements | Income statement, balance sheet and cash flow statement formats, with GAAP presentation and flux-analysis methodology. |
| variance-analysis | Variance decomposition (price/volume, rate/mix), materiality thresholds, narrative generation and waterfall charts. |
| close-management | Month-end close checklist, task sequencing, dependencies, status tracking and common close activities by day. |
| audit-support | SOX 404 control-testing methodology, sample selection, documentation standards and deficiency classification. |
The plugin also ships five slash commands built on top of those skills: /journal-entry, /reconciliation, /income-statement, /variance-analysis and /sox-testing, plus connectors to Snowflake, Databricks, BigQuery, Slack and Microsoft 365. It works inside both Claude Cowork and Claude Code, which is why the walkthrough video can demonstrate it running against a real chart of accounts and a real folder of spreadsheets rather than a sandbox demo.
Set two: the six deal and research skills (Claude for Excel)
The second set, introduced in the October 2025 update, lives inside the Claude for Excel sidebar and targets investment banking, equity research and private equity work rather than corporate accounting:
- Comparable company analysis, producing valuation multiples and operating metrics with a data-refresh capability.
- Discounted cash flow models, including full free-cash-flow projections, WACC calculations, scenario toggles and sensitivity tables.
- Due diligence data packs, converting data-room documents (CIMs, offering memorandums) into structured Excel spreadsheets covering financials, customer lists and contract terms.
- Company teasers and profiles, condensed overviews for pitch books and buyer lists.
- Earnings analyses, extracting metrics, guidance changes and management commentary from quarterly transcripts and filings.
- Initiating coverage reports, combining industry analysis, company deep-dives and valuation frameworks.
These skills, at the time of the October 2025 announcement, were described as in research preview, available exclusively to Claude for Financial Services users on a waitlist, rolling out to Max, Enterprise and Team users first.
Ten agent templates layer on top of both skill sets
Sitting above both skill sets, Anthropic's May 2026 release packaged ten complete agent templates, each combining skills, data connectors and, where useful, specialised subagents into a single ready-to-run workflow. Five cover research and client coverage; five cover finance and operations:
| Research & client coverage | Finance & operations |
|---|---|
| Pitch Builder | Valuation Reviewer |
| Meeting Preparer | General Ledger Reconciler |
| Earnings Reviewer | Month-End Closer |
| Model Builder | Statement Auditor |
| Market Researcher | KYC Screener |
These names overlap conceptually with the six Cowork accounting skills, Month-End Closer clearly draws on close-management, General Ledger Reconciler on reconciliation, but Anthropic has not published documentation confirming the two are literally the same underlying code, and two of the ten templates (Valuation Reviewer, KYC Screener) have no equivalent among the six open-source skills at all. Treat them as a related but separate layer of the same product family, not as a re-statement of the same six skills.
How it works: the Agent Skills architecture
Agent Skills, as a general Anthropic feature, are not finance-specific, they are the same underlying mechanism used for the pre-built PowerPoint, Excel, Word and PDF skills available across claude.ai, the Claude API and Claude Code. Anthropic's own platform documentation describes each Skill as a folder on a virtual machine containing a required SKILL.md file, optional additional markdown references, and optional executable scripts, organised, in Anthropic's own words, "like an onboarding guide you'd create for a new team member."
What makes this efficient at scale, and relevant to why a finance team could plausibly install dozens of Skills without bloating every conversation, is what Anthropic calls progressive disclosure, loading a Skill's content in three stages rather than all at once:
| Level | When it loads | Content |
|---|---|---|
| 1: Metadata | Always, at startup (~100 tokens per Skill) | The Skill's YAML name and description, used to decide whether the Skill is relevant to the current request. |
| 2: Instructions | When the Skill is triggered (under 5,000 tokens) | The body of SKILL.md: procedural workflows, best practices and worked guidance, e.g. reconciliation-item categorisation rules. |
| 3: Resources & code | Only as needed (no cost until accessed) | Bundled reference files and executable scripts, e.g. a chart-of-accounts schema or a variance-waterfall generator, run via bash with only their output entering context. |
Concretely, when a finance user asks Claude to "reconcile this month's bank statement against the GL," Claude matches that request against the description field of the reconciliation Skill, reads the full SKILL.md from the filesystem using bash, and only then, if the task calls for it, opens any bundled reference files or runs any bundled scripts. This is why Anthropic can bundle "no practical limit" of reference content, entire chart-of-accounts documentation, historical close checklists, prior-period workpapers, without every one of those files costing tokens on every single request.
Where these skills actually run differs by surface. In the Claude API, Skills execute inside a sandboxed container with no network access and no runtime package installation. In Claude Code, Skills get the same network access as any other program on the user's machine, which is what lets the finance plugin's connectors reach Snowflake, Databricks, BigQuery or a real accounting system. In claude.ai, network access for custom Skills depends on admin and user settings. This matters directly for a finance deployment: a reconciliation Skill running inside Claude Code against a live ERP connection is architecturally a very different proposition from the same Skill running inside a sandboxed API container.
Availability and pricing
There is no separate "Claude for Finance" subscription. Every layer of this product, the Excel skills, the Cowork plugins, the agent templates, and the open-source accounting-operations plugin, runs on top of Anthropic's standard consumer and business plans. As of 20 August 2026, at a rate of £1 = $1.35 (checked the same day), those plans price out as follows:
| Plan | Price |
|---|---|
| Claude Pro | $20 (£14.80) a month billed monthly; $17 (£12.58) a month billed annually |
| Claude Max | From $100 (£74) a month, across two usage tiers (5x and 20x) |
| Claude Team | $25 (£18.50) a month per standard seat; $125 (£92.50) per premium seat |
| Claude Enterprise | From around $20 (£14.80) a seat, plus usage billed at API rates; requires direct sales contact |
Rollout status also differs skill by skill. The Excel-based comps, DCF, due-diligence and coverage-report skills were described in October 2025 as a research preview, available only to users who had signed up to a waitlist, rolling out to Max, Enterprise and Team users first, with no confirmed general-availability date published since. The open-source Cowork accounting-operations plugin, by contrast, is publicly installable by anyone with a Claude Code or Cowork subscription right now, with no waitlist, via claude plugins add knowledge-work-plugins/finance. The ten May 2026 agent templates are available immediately as Cowork/Code plugins, as Claude Managed Agents (itself in public beta), and as customisable cookbook code. Anthropic has not published separate pricing for any individual Skill, plugin or agent template beyond the underlying plan cost; some data-provider connectors, such as those for FactSet, S&P Capital IQ or Moody's, will carry their own separate licence cost that Anthropic does not itemise or control.
Data, security and governance

Anthropic's finance materials repeatedly stress that these tools are built for human review rather than unattended action. Its May 2026 announcement states directly that "users stay firmly in the loop, reviewing, iterating on, and approving Claude's work" before anything goes to a client, gets filed, or is acted on. The Managed Agents platform adds per-tool permissions, managed credential vaults and full audit logs available in the Claude Console for compliance and engineering teams, and connectors are described as providing "governed, real-time access" to provider data rather than open, unrestricted access. Egnyte, one named connector, is specifically framed around "secure search of permitted data for internal data rooms, investment documents, and approved financial models, while maintaining governed access controls."
Two caveats from Anthropic's own general Agent Skills documentation apply just as much to the finance skills as to any other Skill. First, Agent Skills as a feature are explicitly not covered by Anthropic's zero-data-retention arrangements by default; Skill definitions and execution data follow Anthropic's standard retention policy unless a customer has separately negotiated ZDR coverage. Second, Anthropic's own security guidance is blunt about the risk of using a Skill from an untrusted source: because Skills give Claude new capabilities through instructions and code, "a malicious Skill can direct Claude to invoke tools or execute code in ways that don't match the Skill's stated purpose," potentially leading to data exfiltration or unauthorised system access. Anthropic recommends auditing every file a Skill bundles before installing it and, on Enterprise plans, enabling Skill content scanning for custom Skills uploaded in claude.ai and Claude Cowork, though that scanning does not cover Skills uploaded through the Skills API or Console.
For a finance team, the practical reading is straightforward: the six open-source accounting skills are published by Anthropic itself and are as trustworthy as any first-party Skill can be, but any connector a firm wires up to a live general ledger or a live data room inherits whatever access-control discipline that firm applies to it. Anthropic provides the plumbing for governed access; it does not provide the governance itself.
Early reaction and market context
Independent coverage of the May 2026 agent-templates release was more sceptical than Anthropic's own framing. The Register's report on the launch opened by needling Anthropic's own disclaimer about potential errors, and pointed out that a 64.37% score on Vals AI's Finance Agent benchmark, an improvement Anthropic presented as evidence of state-of-the-art performance, still means the model gets more than a third of tasks on that benchmark wrong, a failure rate the outlet argued financial institutions should weigh carefully before trusting the agents with unsupervised work. It also noted that Anthropic itself frames the ten templates as "reference architectures," developer-facing starting points rather than fully turnkey products, a more modest positioning than "agent" branding might suggest.
Third-party comparisons published through mid-2026 generally position Claude's finance tooling and Microsoft's Copilot agents as solving different problems rather than competing head-to-head on the same axis. Copilot's agents inside Dynamics 365 Finance and Supply Chain Management, such as its Payflow, Procurement and Sales Order agents, run natively within that application's own permissions and business logic, which several 2026 comparison write-ups describe as Copilot's real advantage: integration depth and procurement simplicity for firms whose data already lives inside Excel, SharePoint, Outlook and Dynamics. Claude's finance tooling, by contrast, is pitched as a cross-system reasoning layer, connecting into those same environments via MCP or Office add-ins, with commentators favouring it for long, document-heavy, judgment-driven work such as diligence packs or board narratives that hold an entire filing or data room in context at once. One useful data point on how entangled the two ecosystems have actually become: Claude is now reported to be the default model inside Excel and PowerPoint Copilot in most commercial-cloud regions, with an in-app picker letting users switch between Claude and OpenAI models, a reminder that "Claude vs Copilot" in finance is not always a clean vendor-versus-vendor choice. For a deeper side-by-side of the two ecosystems more broadly, see our full Claude Cowork vs Microsoft Copilot comparison.
Anthropic has not published a formal announcement branded specifically "Claude for Finance" dated around 19 August 2026, despite that date circulating in some secondary coverage. What did happen around that date is narrower and easier to verify: a detailed, well-produced hands-on video demonstrating the open-source Cowork finance plugin, installing it, connecting it to a real accounting workflow, and running its reconciliation, variance-analysis and month-end-close skills against real spreadsheets. That is legitimate, useful, independently produced coverage of a real Anthropic product; it is just coverage of one six-skill plugin from a much larger and older product family, not the debut of a new eight-skill product.
Limitations
- No confirmed 8-skill product. We could not verify a single Anthropic-published set of exactly 8 Agent Skills for finance; 12 real, named skills exist split across two separate 6-skill sets, and we have reported both rather than guessing at a combined figure.
- No August 2026 primary-source announcement. Anthropic's own newsroom shows no dedicated "Claude for Finance" article dated around 19 August 2026; the most recent finance-specific release we could locate on anthropic.com and claude.com is the 5 May 2026 agent-templates post.
- Benchmark accuracy is meaningfully below 100%. Even Anthropic's best-cited result, Opus 4.7's 64.37% on Vals AI's Finance Agent benchmark, leaves substantial room for error on real financial tasks, which is why Anthropic itself frames continuous human review as required, not optional.
- Excel skills remain a gated research preview as of the most recent Anthropic update we could verify, with rollout tied to a waitlist rather than general availability; readers should confirm current access status directly with Anthropic before assuming the comps, DCF or coverage-report skills are open to all paying users.
- Agent Skills are not covered by zero-data-retention by default, a detail buried in Anthropic's general documentation rather than its finance-specific marketing, and one that matters disproportionately for financial-services compliance teams.
- We could not verify specific customer productivity figures. Anthropic names many institutional customers across its finance announcements but does not publish measured time-savings, error-rate reductions or adoption percentages for any of them, so any such figure circulating elsewhere should be treated as unverified.
Who should use it
The six open-source accounting-operations skills are the most immediately useful piece of this product family for most finance teams, precisely because they are publicly installable today, with no waitlist, and because they slot into workflows, reconciliations, month-end close, variance narratives, that already have well-understood human checkpoints. A corporate accounting or FP&A team already running Claude Code or Claude Cowork, with connectors into Snowflake, Databricks, BigQuery or Microsoft 365, can reasonably pilot this plugin on a single close cycle before deciding whether to extend it.
The Excel-based investment-banking skills are a better fit for equity research, M&A and private-equity teams already inside the Claude for Financial Services ecosystem, but given their research-preview, waitlist-gated status at the time of the most recent verifiable Anthropic update, teams outside that programme should treat them as a roadmap item to watch rather than something to plan a workflow around this quarter.
Teams whose entire financial stack already lives inside Dynamics 365, Excel, SharePoint and Outlook, with no appetite to introduce a second AI vendor, are probably better served starting with Microsoft Copilot's native Dynamics agents, per the third-party comparisons cited above, and revisiting Claude's tooling if and when a task genuinely needs long-document, cross-system reasoning that Copilot's narrower, application-bound agents are not built for.
The bottom line
Claude for Finance is real, useful, and has been built up carefully over more than a year, but it is not the single, freshly minted "8 skills" product that recent framing suggests. What actually exists, verified directly against Anthropic's own newsroom, its solutions pages, its GitHub repository and its Agent Skills documentation, is a two-track family of Agent Skills, six for accounting operations and six for investment-banking research, sitting underneath a broader set of Cowork plugins and agent templates, all running on Anthropic's standard subscription plans rather than a bespoke finance price. For teams evaluating it, the open-source accounting plugin is the fastest, lowest-risk place to start; everything else is worth watching, but worth confirming current availability on directly before building a workflow around it.
This article was built from Anthropic's own published newsroom articles (July 2025, October 2025, February 2026 and May 2026), its claude.com solutions and pricing pages, its public anthropics/knowledge-work-plugins GitHub repository, and its official Agent Skills platform documentation, cross-checked against independent reporting from The Register and third-party finance-tooling comparisons, plus the 19 August 2026 hands-on video that prompted this piece.
Last updated: 20 August 2026. This article will be revised if Anthropic publishes a dedicated "Claude for Finance" announcement, confirms an 8-skill set, or moves the Excel-based skills out of research preview.
Get the free guide: Claude vs ChatGPT, Gemini & Grok
A 20-page playbook covering everything you need to choose and use the big four AI models in 2026, full cost and feature comparisons, what each is best (and worst) at, and how-tos for images, vectors, building a website, Claude Code and more.






