GPT-6 Astra vs Claude Fable 5.1: QBR Decks (2026)

A quarterly review deck is a knowledge-work packet, not a chatbot transcript. GPT-6 Astra and Claude Fable 5.1 are the two frontier models an RIA can point at that packet. Fable 5.1 shipped 1 September 2026 and is live on paid Claude, API, and clouds. Astra shipped 3 September 2026 and is not generally on ChatGPT that day. This page is only those two products.

AA-Briefcase is the independent lab's long knowledge-work eval: multi-week professional packets scored on analysis and on presentation. Fable 5.1 leads the published absolute Elo. Astra gained versus OpenAI's previous flagship on the same family of work, with presentation Elo down. The advisor decision is which failure mode you will supervise: thin analysis, or a pretty deck that over-polishes the wrong story.

TL;DR

• Pick Claude Fable 5.1 when the job is a long QBR packet and you want the published AA-Briefcase leader at 1,694 Elo, with analytical quality Elo 2,025 and Intelligence Index 66 at max.

• Pick GPT-6 Astra when the org can call gpt-6-astra, you want AA Intelligence cost per task of $1.67 instead of $3.69, and you will accept that Astra's Briefcase presentation Elo fell versus OpenAI's previous flagship.

• List price is a tie at $10 input / $50 output per 1M tokens. Cache is not: Astra cache reads $1, Fable cache reads $0.25.

• Neither model should publish a client deck without a CFP pass on the numbers and a separate pass on the slides.

A day in the life of a financial operator

QBR week in an RIA ops seat is a stack of household files, a shared drive of last quarter's decks, and a calendar that does not slip. Fourteen reviews sit on the week. Each pack needs performance, allocation, tax-lot notes, and a next-step slide. The operator is not designing a brand. The operator is trying to leave the conference room without a client catching a wrong AUM or a chart that implies a promise the IPS does not make.

The old motion is a paraplanner in PowerPoint or Google Slides from 7 a.m. to 7 p.m. The new motion is a model that swallows the household export and returns a 12- to 18-slide draft. That is the Briefcase-shaped job: long, professional, scored on whether the analysis is right and whether the artifact looks like something you can put on a TV in a conference room.

Fable 5.1 is the model you can actually open this afternoon on a paid Claude seat. Astra is the model you can open if the org is already in Trusted Access, limited-org, or Foundry Limited Access, or if an admin has enabled Enterprise. A day in the life that assumes a ChatGPT tile named GPT-6 Pro is a day that has not read the 3 September access note.

Proposal software still owns e-sign and the pretty template. The model owns the first narrative. Keep those layers straight; see advisor proposal software when the gap is the packet, not the prose.

The workflow, mapped

The packet starts as a meeting on the books, becomes a folder of exports, becomes a draft deck, becomes a reviewed deck, becomes the room. Wealthbox or Salesforce can be the household system of record. Google Slides or PowerPoint can be the artifact. The model sits in the middle only if someone owns the hold.

Worked example: after the first-meeting handoff lands, the ops seat exports the household and calls Google Slides presentations.batchUpdate, documented in Google's Slides API, to drop 16 slides into the QBR template. US Tech Automations triggers that write after a CRM stage change, sends the export to Claude Fable 5.1 or GPT-6 Astra at the shared $10 per 1M input list, and holds the 16-slide deck in a review queue so a CFP can check 9 allocation charts before the client link goes out. Three figures are load-bearing: 16 slides, $10 list, 9 charts. The API token is load-bearing because a "make me a deck" chat that never writes presentations.batchUpdate is not a workflow.

Onboarding to first meeting is the same shape with a thinner packet. If the firm still builds that path in email, fix the handoff first: new client onboarding to first meeting.

Fable 5.1 thinking is adaptive and on by default. Editing earlier turns invalidates thinking blocks, and earlier Claude models cannot read Fable 5.1 thinking. Astra has no none reasoning and needs the Responses API for tools. Those are build constraints, not slide taste.

What it costs to keep doing it manually

Manual QBR decks cost senior hours, not tokens. A 16-slide pack that takes a paraplanner three hours is a staffing line. A model draft that still needs 40 minutes of CFP cleanup is a different staffing line. The wrong comparison is "tokens versus zero." The right comparison is hours still on the calendar after the draft lands.

according to the Investment Company Institute in its 2024 Investment Company Fact Book, U.S. registered investment company total net assets were $33.6 trillion at year-end 2023. That is the ocean the QBR is trying to explain to one household. The deck is a communication control, not a design contest.

Hour totals are a workflow sizing example at stated rates, not a salary survey. Use your loaded cost. Token prices in the next tables are the vendor list as of 2026-09-03.

A CRM that does not hold the household will make every one of those hours worse. If the firm is still arguing Wealthbox versus Salesforce as the file, settle that before you pick a model: Wealthbox vs Salesforce for advisors.

How we evaluated

Weights assume a QBR packet with a human hold. A team that only wants cheaper tokens should invert cost and Briefcase. A team that presents on a TV in the conference room should keep presentation Elo visible even when analysis Elo is the headline.

according to Artificial Analysis, Claude Fable 5.1 leads AA-Briefcase at 1,694 Elo (+122 versus Fable 5), with analytical quality Elo 2,025 and presentation Elo 1,495. The AA Fable Intelligence run used Anthropic's default safety fallback, with about 4% of output tokens routed to Opus. Do not pretend the 66 is a pure Fable-only lab.

The tool comparison

Same list sticker. Different cache, different task $, different access, different Briefcase story. Fable 5.1 is live. Astra is limited. Fast mode is 2× Standard on OpenAI API docs and 2.5× on the Help Center Codex/Work card; Fable does not use that OpenAI Fast SKU.

AA articles 2026-09-01 and 2026-09-03 plus leaderboard 2026-09-03; OpenAI launch table 2026-09-03; Anthropic and OpenAI list prices 2026-09-03. OpenAI's table is provider-run.

according to Anthropic, Fable 5.1 is available on Claude Pro, Max, Team, and Enterprise, plus API and major clouds, at $10 / $50 list with cache reads at $0.25. Mythos 5.1 is the invite-only twin with looser cyber and life-sci gates; it is not a public picker option and it is not this page's second product.

according to Microsoft, GPT-6 Astra on Foundry Standard Global short context is $10 input / $1 cached / $12.50 cache writes / $50 output per 1M tokens. That is the Astra list, not a discount to Fable's cache.

Payback math

You do not pay back a QBR model by "saving 3 hours" on a slide that still ships a wrong allocation. You pay back by cutting paraplanner draft hours while keeping two human passes. Token $ is the small line. Fable is more expensive per AA Intelligence task. Astra is more expensive per cache-light, cache-read-heavy Claude loop is where Fable's $0.25 hits.

according to The Verge, Anthropic cut Fable 5.1 cache reads 75% versus the prior $1 cache-read price, leaving list I/O unchanged. That 75% is why a cached QBR template favors Fable even though AA task $ favors Astra.

Task $ is AA's Intelligence cost/task, not a QBR invoice. Use it to rank burn, then measure your own packet tokens. Do not say Fable is cheaper than Astra on the AA task column; it is not.

Fable 5.1 AA-Briefcase Elo: 1,694

Astra AA Intelligence cost/task: $1.67

Fable cache reads: $0.25 per 1M

Who this is for

This page is for RIA ops leads, lead advisors, and marketing-ops partners who already run a CRM and a slide template, and who want a model to draft the QBR packet. Firm size is "we present every quarter," not a headcount cutoff. Current stack is a household CRM plus Google Slides or PowerPoint plus a human who will still own the numbers.

Red flags: no template (the model will invent a brand); no CFP numbers pass; waiting for Astra's ChatGPT tile before doing any QBR automation; treating Mythos or a cyber waitlist as a slide model.

When NOT to use US Tech Automations: if one advisor builds 20 slides by hand in Google Slides and never reuses the structure, stay in the native editor. If Zapier, Make, or n8n already copies a closed-won meeting into a shared folder and that is the only seam, keep that copy. If Wealthbox or Salesforce already generates the report you present, do not add a second deck factory. Add orchestration when presentations.batchUpdate (or the PowerPoint equivalent) must run after a CRM event, call a named model, and sit in a review queue.

Pros and cons

GPT-6 Astra

Pros

• AA Intelligence cost per task at max is $1.67, less than Fable's $3.69 on the same column.

• Briefcase Elo improved about 80 points versus OpenAI's previous flagship, so packet work is not a regression on analysis.

• List $10 / $50 matches Fable, so finance can compare cache and task $ instead of arguing stickers.

• 1.05M context and 128K max output on the Astra card, enough for a household export plus speaker notes.

Cons

• Not generally on ChatGPT on 3 September 2026; Enterprise off until an admin enables it.

• Briefcase presentation Elo is down versus OpenAI's previous flagship, which is the wrong direction for a TV in a conference room.

• Absolute AA-Briefcase Elo was not published for Astra on 3 September 2026, so you cannot rank it on the same 1,694 scale without a later AA print.

• Cache reads are $1 versus Fable's $0.25, which hurts a templated QBR loop.

Claude Fable 5.1

Pros

• Published AA-Briefcase leader at 1,694 Elo, with analytical quality Elo 2,025.

• Independent Intelligence Index 66 at max versus Astra's 61.

• Live on paid Claude, API, and clouds on 1 September 2026.

• Cache reads $0.25 per 1M, a 75% cut versus the prior $1 cache-read price.

Cons

• AA Intelligence cost per task at max is $3.69, more than Astra's $1.67.

• Presentation Elo 1,495 trails its own analytical quality 2,025; plan a human polish pass.

• AA's 66 used a safety fallback with about 4% of output tokens routed off Fable.

• Forced tool_choice any/tool returns 400; thinking blocks break if you edit earlier turns.

FAQs

Which model wins AA-Briefcase for advisor slide packets?

Claude Fable 5.1. according to Artificial Analysis, Fable 5.1 is at 1,694 Elo on AA-Briefcase. Astra's absolute Elo was not on that print; the Astra article reports about +80 Elo versus OpenAI's previous flagship and a drop in presentation Elo.

Is GPT-6 Astra cheaper for QBR drafts?

On the AA Intelligence task column, yes: $1.67 versus $3.69. On cache-heavy template loops, Fable's $0.25 cache read can invert the bill. List I/O is a tie at $10 / $50.

Can we run Astra from ChatGPT for next week's reviews?

Not as a general ChatGPT model on 3 September 2026. Limited orgs, Trusted Access, and Foundry Limited Access are the door. Fable 5.1 is the model a paid Claude seat can run this afternoon.

Do we skip the CFP pass if Briefcase Elo is high?

No. Fable's presentation Elo is 1,495 against analytical quality 2,025. Astra's presentation Elo moved the wrong way versus OpenAI's previous flagship. The hold stays.

Should Zapier build the deck?

Zapier, Make, or n8n can drop a file in a folder after a meeting is booked. That is a copy. A QBR packet that calls a model, writes slides, and waits for a CFP is a workflow with a hold. Use the copy tools when copy is the whole job.

What about METR time horizon for these packets?

Not published for either model as of 3 September 2026. Do not invent hour numbers. Use Briefcase, GDPval, and your own review-hour log.

Key Takeaways

• Fable 5.1 is the published AA-Briefcase leader at 1,694 Elo; Astra gained about +80 versus OpenAI's previous flagship and lost presentation Elo on that family of work.

• Intelligence Index at max is Fable 66 versus Astra 61. Task $ at max is Astra $1.67 versus Fable $3.69. List is $10 / $50 both ways.

• Fable is live. Astra is not generally on ChatGPT on 3 September 2026.

• QBR decks still need a numbers pass and a polish pass. Elo is not a send button.

• US Tech Automations belongs when a CRM event must call a named model, write slides, and hold for a CFP. Native chat and a single Zapier/Make/n8n copy do not.

Pick Fable 5.1 for the packet you can run today and for the published Briefcase leader. Pick Astra when the org can call it and the cost owner is living on the AA task column. Route the hold through US Tech Automations only when the deck is a workflow, not a chat; the shape is on agentic workflows.