I Tested Fable 5.1 vs Fable 5 vs Opus 5 (Cost/Speed/Design)

Summary
In this video, presenter Brock Mesarich conducts a hands-on benchmark comparing Anthropic's Claude Fable 5.1 against Claude Fable 5, Claude Opus 5, and OpenAI's Codex Sol / Terra models. He evaluates each model across three effort tiers (Low, High, and Max) on the same multi-step task: generating photorealistic SpaceX Falcon 9 videos using a Higgsfield MCP connector and coding an animated interactive landing page.

What is shown

• [00:31] Introduction of the benchmark scorecard tracking effort levels (Low, High, Max), visual design score out of 10, generation run time, and token/API cost.
• [01:13] Explanation of prompt caching mechanics and Anthropic pricing differentials between standard input tokens ($10/M tokens) versus cached input tokens ($0.25/M tokens on Fable 5.1 vs. $1.00/M tokens on Fable 5).
• [02:28] Navigating the Claude Desktop app interface to configure models and adding the Higgsfield MCP connector (https://mcp.higgsfield.ai/mcp) via the custom connectors menu.
• [04:10] Prompting Claude Opus 5 with the Higgsfield connector to produce five 1080p photorealistic Falcon 9 clips using the Seedance 2.5 video generation model, then reviewing the generated outputs at [05:18].
• [05:53] Prompting each model variant across Claude and ChatGPT with the identical prompt to build an animated Falcon 9 landing page utilizing the generated video clips.
• [07:33] – [17:58] A blind evaluation of the generated websites, reviewing layout, animations, countdown timers, and visual styling:
  • Website 1 (Opus 5 Max): 6/10 look rating, 17m 57s active time, $10.70 cost [08:55].
  • Website 2 (Opus 5 High): 5/10 look rating, 26m 06s active time, $11.01 cost [10:02].
  • Website 3 (Fable 5 Max): 6/10 look rating, 26m 41s active time, $24.01 cost [10:47].
  • Website 4 (Codex 5.6 Terra light): 7/10 look rating, 10m 53s runtime, cost N/A [12:04].
  • Website 5 (Fable 5.1 High): 8/10 look rating, 18m 05s active time, $8.78 cost [13:24].
  • Website 7 (Codex 5.6 Sol High): 6/10 look rating, 13m 07s runtime, cost N/A [14:48].
  • Website 8 (Fable 5.1 Max): 7/10 look rating, 26m 55s active time, $10.67 cost [15:37].
  • Website 9 (Opus 5 Low): 2/10 look rating, 10m 16s active time, $6.58 cost [16:27].
  • Website 10 (Fable 5 High): 6/10 look rating, 2m 11s active time, $5.78 cost [17:08].
  • Website 12 (Fable 5 Low): 7/10 look rating, 5m 04s active time, $9.76 cost [17:59].
• [18:13] – [20:20] Presentation of the completed scorecard and rankings sorted by visual quality (top: Fable 5.1 High) and cost (cheapest: Fable 5 High at $5.78; most expensive: Fable 5 Max at $24.01).

Claims & numbers

• The presenter notes Anthropic announced Claude Fable 5.1 and Claude Mythos 5.1 on September 1, 2026 [00:01].
• The presenter cites Anthropic benchmark numbers showing Fable 5.1 achieving 52.6% on Terminal-Bench-Science 0.1, 55.8% on Terminal-Bench 4.0, 77.9% on OSWorld 2.0 (partial), and 31.2% on AutomationBench [00:27].
• The presenter claims prompt caching read costs dropped 75% from Fable 5 ($1.00 per million tokens) to Fable 5.1 ($0.25 per million tokens), while uncached input tokens remain at $10.00 per million tokens [01:40].
• Higgsfield MCP charged 72 credits per video (360 total for 5 videos) via Seedance 2.5 [05:01].
• Fable 5.1 High produced the presenter's top-rated website (8/10) at a session cost of $8.78 and 18m 05s active runtime [13:35].
• The most expensive run was Fable 5 Max at $24.01 and 26m 41s runtime [11:23], whereas Fable 5.1 Max cost $10.67 with 26.9 minutes of wall clock time [15:48].
• Fable 5 High was the cheapest run recorded in Claude at $5.78, taking only 2 minutes and 11 seconds [17:11].

Notable quotes

• [01:29] "Think of caching like a bookmark that we are able to give an AI."
• [13:48] "If we're learning anything here, at least for me, it's that sometimes a model doesn't necessarily matter that we are using."
• [20:46] "Using a model like Fable 5.1 Max at the highest effort level is probably overboard for whatever it is you're trying to do."

Assessment
This is an authentic, independent third-party user review and empirical testing video comparing frontier models in Claude Desktop and ChatGPT. The presenter shows real screen captures of the workflows, command outputs, session billing metadata, and the resulting websites without deceptive staging.

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