In April, the highest-scoring AI model you could buy was Claude Opus 4.7, at $10 per million tokens. According to Artificial Analysis data, in October Anthropic released Claude Haiku 5.5, which rises from 40.7 to 43.4 on the Intelligence Index and costs 20 cents: a 50-fold cut in five and a half months.
The overall picture: once a score is reached, the price of matching it collapses within months. The price of the top model, by contrast, stays flat. Artificial Analysis tracked ten models that took the top score between September 2024 and September 2025, comparing each launch price with the cheapest model matching that score, day by day. In the first three months the typical top score became about four times cheaper. At six months a model with the same score typically cost twenty times less; after a year, about 105 times less.
Waiting about six months
Since September 2024, twenty-one models have taken the top score. For seventeen of them, a sub-dollar model has eventually matched it. The wait ranged from two and a half to eight and a half months, with a median of 175 days. Those matching models were mostly Chinese open-weight models such as MiniMax, DeepSeek and Z.ai, alongside small models from xAI and OpenAI.
The four newest top models are still waiting, all from Anthropic. The oldest, Claude Fable 5, has held its score for 131 days, within the usual range. Today's best sub-dollar model, Xiaomi's MiMo-V2.6-Pro, scores 46.3 against Fable 5's 49.6: on the usual timetable, a sub-dollar model should match it around December.
The price of the strongest model does not follow the same curve. It sat near $30 in 2024; since then it has swung between $3.44 and $35 with no clear trend. The typical top model cost $6 in 2025 and $10 in 2026. Claude Opus 5.5, currently at the top, costs $8.
The launch price is not the final price
The launch price is often the highest a model will ever carry. OpenAI cut o3's price by 80% two months after launch, and GPT-5.6 Luna's by 80% three weeks after. DeepSeek cut V4 Pro by 75% within a month. The charts use each day's price, so these cuts appear as steps.
Price per token is not the whole story
Prices are per token, and models don't use the same number of tokens. Some reason at length before answering, and that reasoning is billed. ModelCompare ran the same 163 conversations from its hallucination tests through fifteen models and summed what each cost. Claude Haiku 5.5 costs 40 times less than Claude Opus 5.5 per token and 17 times less per conversation. Gemini 3.8 Flash, at $1.50 per million tokens against $4 for GPT-6.1 Sol, costs about the same as GPT-6.1 Sol per conversation. The only way to know what you will actually spend is to run your own prompts.
The Eiffel Tower case
A practical comparison shows the gap too. According to Crypto Briefing, Claude Opus 5.5 took two hours and $33.45 to build the Eiffel Tower, while Gemini 4 Argon finished it in one shot for $0.23, roughly 145 times cheaper.
No comments yet. Be the first!