Token pricing for GPT models

Hey there!

Welcome back to The Pulse, where we dive into interesting AI stories and trends backed by data, all presented through simple visuals.

> GPT 4.1 models released exclusively in API today

> gives high performance at low costs

> up to 1M tokens of context + excellent at coding

> largely outperforms 4o models across the board & even GPT 4.5 in some benchmarks (Aider's polyglot, AIME '24)

> beats o1 (high), o3-mini (high) & GPT-4.5 at SWE-bench verified

> U.S. federal level: 171 AI bills proposed in 2023 -> 221 in 2024:

• nearly tripled since 2022, yet only 4 passed

• indicates regulatory lag while AI incidents rose 56.4% over 2023

> U.S. state level: dramatic growth in laws passed - total 1 bill (2016) to 49 (2023) then >2x to 131 (2024):

• California: 42 total (22 in 2024)

• Utah: 17 total (12 in 2024)

• New York: 8 total (5 in 2024)

> International (2024):

• Russia: 7 bills

• Portugal: 5 bills

• Belgium: 5 bills

> China: despite ranking #2 in AI, passed only 1 bill in 2024 (4 total since 2016), pointing to safety concerns

> open models catching up to closed models in performance:

• Jan 2024: best closed model beat best open model by 8%

• Feb 2025: only outperforming by 1.7%

> same trend follows on other benchmarks such as MMLU, HumanEval, MATH, etc.

> current best open model (acc. to Artificial Analysis) DeepSeek R1 only 8 pts behind best closed model (Gemini 2.5) on the intelligence index