Claude Haiku 5.5 costs 12x more than GPT-6 Luna on the same voxel pagoda test
Fun-Meaning-6474 · reddit · 2026-10-08
A Reddit user benchmarked the newly released Claude Haiku 5.5 against GPT-6 Luna on the voxel pagoda test and found a massive cost gap:
- Usage: at xhigh effort, Haiku 5.5 burned 268.8M input tokens (4.46M output) vs Luna's 64.5M input (1.18M output)
- Cost: API-equivalent cost was $24.96 for Haiku vs $1.96 for Luna — roughly 12x
- Why: Haiku 5.5 is a token goblin, and its API pricing quintuples past 100k context, so nearly every turn in an agent session bills at 5x; Luna's 5x threshold sits at 272k, where it auto-compacts anyway
Tests were run through subscriptions on atomic.chat.
Related event: Claude Haiku 5.5 ranks 2nd in intelligence but real costs spiral(16 posts)→
More from Models
- Same $0.10/$0.50 price, 40% different bills: GPT-6 Luna beats Haiku 5.5 in real billing test — aakashgupta · 2026-10-10
- GPU kernel bench leaks: DeepSeek v4.1 flash beats GLM-5.3 and Haiku 5.5 — teortaxesTex · 2026-10-10
- Step 5 Preview makes clean slides, Qwen goes report-style in Apple deck showdown — CodeByPoonam · 2026-10-10
- The key test: editing one number to see if each model's charts update correctly — CodeByPoonam · 2026-10-10
- Blogger benchmarks StepFun Preview vs Qwen 3.8 Max on Apple FY2025 analysis deck — CodeByPoonam · 2026-10-10
- User says Grok bot's hidden subagents and instant replies ruin other LLM experiences — rudrank · 2026-10-10