ai / llm benchmarks / algorithmic trading
Claude Haiku 5.5 Cost Me 13x More Than Muse Spark
Anthropic's newest budget model, Claude Haiku 5.5, launched at ten cents per million input tokens, so Austin tested it on the real prompts behind his AI trading agent. Once a prompt passes 100,000 tokens, Anthropic charges five times more, and his agent's prompts run about 180,000 tokens, so every call paid the higher rate.
Two hundred seven decisions cost $19.86 on Haiku. Meta's Muse Spark 1.3 ran the same 207 decisions for $1.48. Haiku was much faster, at a median of four seconds per decision instead of twenty, and it did win one job: on his stock screener it answered 19 of 22 questions, so that slot switches to Haiku.
Transcript
0:00Anthropic's newest budget model cost me 13 times more than Meta's Muse Spark.
0:04Claude Haiku 5.5 came out this week at $0.10 per million input tokens, so I tested it on the real prompts behind my AI trading agent.
0:13Once a prompt passes 100,000 tokens, Anthropic charges 5 times more.
0:18My agent's prompts run about 180,000 tokens, so every call paid the higher rate.
0:23207 decisions cost $19.86.
0:27Meta's Muse Spark 1.3 ran the same 207 decisions for $1.48.
0:33Haiku was much faster, at a median of 4 seconds per decision instead of 20.
0:37It did win one job.
0:40On my stock screener, Haiku answered 19 of 22 questions, so that slot switches today.
Join the conversation
Loading conversation…