I Spent 24 Hours Testing Haiku 5.5: Here Are The Results

Summary
In this video, presenter Nick Saraev details an empirical evaluation comparing Anthropic’s Claude Haiku 5.5 against Claude Sonnet 5.5, Claude Opus 5.5, and Claude Fable 5.1 across multiple complex generation tasks. Saraev spent over $700 running identical single-shot prompts to build 3D Three.js browser games, web apps, interactive landing pages, SVG animation documentaries, and data audits to evaluate quality, generation speed, and API cost.

What is shown

• Benchmark Dashboard Overview [00:23]: Saraev shows a web dashboard tracking round-by-round builds, total costs, execution times, token counts, and prompts across tested models.
• MOBA Game Demo (League of Legends homage) [01:11]: Side-by-side gameplay of Three.js MOBAs generated by Haiku 5.5 (Riftbound), Sonnet 5.5 (Aetherfall), and Opus 5.5 (Spirefall).
• Narrative Game (Breaking Bad homage) [04:27]: Visual and interactive comparison of episodic crime games: Haiku 5.5’s Halcyon ($8.87, 104.28M tokens), Sonnet 5.5’s Nobody ($40.29, 145.04M tokens), and Opus 5.5 ($41.42, 91.81M tokens).
• 3D Luxury Watch Launch Site [08:28]: Interactive product landing pages featuring scroll-triggered 3D exploded watch assemblies (Horizon Squelette) produced by Haiku 5.5 ($0.37), Sonnet 5.5 ($2.88), and Opus 5.5 ($5.45).
• Forensic Fraud Audit [10:44]: An accounting challenge analyzing 1,788 payments for 8 planted fraud schemes; all three models accurately flagged 8/8 schemes with 0 false positives, with Haiku costing $0.41 versus Opus at $1.92.
• FPS Game (Modern Warfare homage) [12:27]: Playable browser-based FPS arena shooters generated by Haiku 5.5 (Gray Lantern), Sonnet 5.5 (Iron Tide), and Opus 5.5 (Black Tide).
• Animated Documentary Video [15:38]: Code-synthesized SVG animation documentary on the history of flight generated by Haiku 5.5 ($0.55), Sonnet 5.5 ($10.74), and Opus 5.5 ($7.82).
• Voxel Sandbox (Minecraft homage) [18:07]: Three.js procedural voxel world generation compared between Haiku 5.5 ($0.17), Sonnet 5.5 ($2.08), and Opus 5.5 (Blockland, $5.14).
• Personal Finance SaaS Dashboard (Ledgerly) [20:20]: Interactive personal finance web apps with charts, transactions tables, dark/light themes, and CSV export.
• Hack-and-Slash Game (God of War homage) [23:00]: Third-person Greek myth combat games: Haiku 5.5’s Ashes of the Oath ($9.67), Sonnet 5.5’s Ashen Wrath ($34.31), and Opus 5.5’s Ashes of Sparta ($41.99).
• Attack on Titan / Minecraft Hybrid (Attack on TitanCraft) [25:34]: 3D aerial grappling/swinging mechanics tested on Haiku 5.5 ($5.15), Sonnet 5.5 ($23.78), Opus 5.5 ($24.83), and Fable 5.1 ($23.35).
• Pricing & Subagent Architecture [29:15]: Bar chart breakdowns showing token pricing, prompt caching tiers, the 100K token cliff, and a diagram illustrating Opus 5.5 planning while orchestrating Haiku 5.5 subagents.

Claims & numbers

• The presenter says he spent over $700 on real API bills split-testing Claude Haiku 5.5 against Sonnet 5.5 and Opus 5.5 across identical prompts [00:01].
• For the forensic accounting audit, the presenter reports that Haiku 5.5, Sonnet 5.5, and Opus 5.5 all achieved a perfect score of 8/8 planted fraud schemes identified ($391,303 total) with 0 false flags [11:09].
• The presenter states standard API pricing per million tokens (input/output) as: Fable 5.1 at $10 / $50; Opus 5.5 at $4 / $20; Sonnet 5.5 at $2 / $10; Haiku 4.5 at $1 / $5; and Haiku 5.5 at $0.10 / $0.50 [29:19].
• The presenter notes Haiku 5.5 is 10x cheaper than Haiku 4.5 and 40x cheaper than Opus 5.5 [29:36].
• The presenter explains the “100K-token price cliff”: prompts up to 100K tokens cost $0.10 in / $0.50 out per million tokens on Haiku 5.5, but past 100K tokens, pricing increases 5x to $0.50 in / $2.50 out, while prompt cache reads cost one-tenth ($0.01 / $0.05) [30:10].

Notable quotes

• “I just spent over $700 split testing Claude Haiku 5.5 across other Claude models like Sonnet and Opus 5.5...” [00:00]
• “I wouldn’t say the generations of Opus were... 40x better than the generations of Haiku. In fact, I would say the gap there is a lot smaller...” [29:56]
• “Opus 5.5 or another frontier model plans the work. Haiku is what sort of is your fingers and does the job...” [1:53]

Assessment
This is a hands-on independent review and empirical benchmarking video demonstrating full, functional browser artifacts created directly from LLM API calls. All software artifacts and gameplay sessions are shown running live in-browser with complete pricing and token usage metrics recorded on screen.

Described by gemini-3.8-flash on 2026-10-12 from the video’s audio and frames.