Claude Fable 5: Better Than Opus 4.8?

Summary
Jamie Keet from Teacher's Tech presents an independent hands-on evaluation of Anthropic's Claude Fable 5, comparing it head-to-head against Claude Opus 4.8. Through four practical business tests—analyzing charts in PDFs, auditing spreadsheet calculations, synthesizing multi-file launch memos, and testing domain guardrails—he assesses whether Fable 5's capabilities justify its double pricing tier.

What is shown

• [00:53] Architecture breakdown: Diagram explaining the "Mythos Class" foundation, contrasting restricted access to Mythos 5 with the safeguarded, publicly accessible Fable 5.
• [01:28] Interface & plan timeline: Claude model dropdown showing Fable 5, Opus 4.8, Sonnet 4.6, and Haiku 4.5, alongside a timeline for free inclusion versus credit-based usage.
• [02:25] Test 1 (PDF visual vs. text discrepancy): Side-by-side test in incognito mode with an unlabelled bar chart contradicting the written paragraph; both models catch the conflict, while Fable provides deeper longitudinal analysis.
• [04:30] Test 2 (Spreadsheet analysis & error detection): Upload of basecamp-brew-sales-mar25-feb26.xlsx. Neither model flags an unprompted formula error, but upon direct query at [06:13], both identify a July 2025 digit-swap error ($8,820 vs. $8,280), with Opus running verification code.
• [07:44] Test 3 (Multi-document synthesis): Five mixed files (notes, emails, spreadsheet, PDF) ingested to draft an executive memo. Both identify key conflicts, but Fable 5 makes an arithmetic error summing unit sales (3,690 vs. 4,090).
• [10:24] Test 4 (Safeguard rerouting): A benign question on coffee roasting chemistry triggers Fable 5's automated safety guardrail, silently delegating the prompt to Opus 4.8.
• [11:50] Evaluation scorecard: Jamie summarizes comparative performance versus the 2x token pricing.

Claims & numbers

• The presenter says Fable 5 is the first model released in Anthropic's Claude 5 family, sharing underlying weights with the enterprise-gated Mythos 5.
• The presenter says Fable 5 was included with paid Claude subscriptions at no extra cost through June 22, 2026, before requiring usage credits starting June 23, 2026.
• The presenter states API token pricing for Fable 5 is $10 per million input tokens and $50 per million output tokens—exactly double Opus 4.8 ($5 / $25 per million tokens).
• The presenter cites analytics partner Hex claiming Fable 5 is the first model to score above 90% on their benchmark of long-running analytical tasks (10 points ahead of Opus).
• The presenter states that Fable 5's safety mechanisms automatically reroute queries regarding cybersecurity, biology, chemistry, and model distillation to Opus 4.8, affecting fewer than 5% of all chats.
• The presenter notes that Fable 5 logs have a 30-day retention policy for safety monitoring, though Anthropic confirms this data is not used for model training.

Notable quotes

• "Anthropic's own framing is that the longer the task, the bigger its lead over the other models." [00:45]
• "The cheaper model wants to fix your document, and the pricier one wants to tell you what it means." [04:14]
• "Every trap got caught by both models, every time. The differences came down to one extra sentence here, one sharper question there..." [11:56]

Assessment
This is an authentic, independent review and real product demonstration rather than marketing hype. The presenter executes tests in incognito chats to eliminate conversational memory bias and highlights genuine limitations, such as Fable 5's oversensitive safety rerouting and a mathematical summation error.

Described by gemini-3.8-flash on 2026-09-29 from the video's audio and frames.