$/M tokens is broken: Opus 5.5 costs ~6x Sonnet per task at similar scores
lordmairtis · reddit · 2026-10-12
A Reddit user argues that $/M-token pricing is a misleading way to compare models, citing Artificial Analysis' Cost per Intelligence Index Task chart: GPT-6 Luna at $0.07 per task, GPT-6.1 Sol at $0.72, Claude Sonnet 5 (max) at $5.09, Claude Opus 5 (max) at $5.86, and Claude Opus 5.5 (max) at $5.98.
Key points:
- Opus 5.5 has double the $/MT price of Sonnet 5.5 with near-identical benchmark scores, yet far higher real per-task cost
- GPT-6.1 Sol scores slightly lower at the same $/MT but costs a fraction per task
- Large models eating limits on simple prompts forces users to build proxies and routers — and then verify whether the routed model is actually cheaper
- Takeaway: run a fixed set of simple tasks on every new model to measure real cost/benefit, with no guarantee it reflects real workloads
More from Models
- Gadsby-4B Hits Hugging Face: Constrained Decoding Turns Gemma 3 4B Into a Lipogram Model — cephaloform · 2026-10-12
- 50 AI Models Have Played Cowcraft, an Online Limit-Testing Arena for LLMs — djcows · 2026-10-12
- After days of testing, dev ditches sol 6.1 and astra 6 for agentic work, returns to Opus — NERDDISCO · 2026-10-12
- Mathematicians push back on OpenAI's model-generated proofs — ZeeshanZiaML · 2026-10-12
- Anthropic's Claude can't even do substring search in chat history, user shows — AaronBergman18 · 2026-10-12
- Every Recent Claude Loves to Say Things Are 'Carried' or 'Held' — repligate · 2026-10-12