Karpathy tours the practical differences between the major models as a daily user; great intuition-builder.
How I use LLMs
Andrej Karpathy Feb 2025
Watch on YouTube youtube.com →There is no best model, only a best model for your task at your price: test the top two or three on your own real examples rather than trusting benchmark headlines. Design your code so the model is a swappable config value (a thin gateway or SDK abstraction does this), because prices and leaders change every few months. Open-source models make sense once you have scale, strict data constraints, or a narrow task; for most first features a hosted API wins on speed to ship.
17 resources.
Karpathy tours the practical differences between the major models as a daily user; great intuition-builder.
Andrej Karpathy Feb 2025
Watch on YouTube youtube.com →YC partners on how model progress should shape what founders build now versus wait on.
Y Combinator (Lightcone Podcast) 2024
Open creators.spotify.com →A structured comparison across reasoning, coding, context, and compliance rather than hype.
IntuitionLabs 2026
Open intuitionlabs.ai →Side-by-side output quality on real writing tasks, the kind of eval founders should copy.
Type.ai 2025
Open blog.type.ai →Explains what each benchmark measures so you stop choosing models on a single headline number.
Future AGI 2025
Open futureagi.com →A serving-infrastructure company's practical map of the open-weight landscape.
BentoML 2026
Open bentoml.com →A decision framework for the open-vs-closed call, written for teams shipping to customers.
Microsoft 2025
Open techcommunity.microsoft.com →Covers the hybrid pattern most teams actually land on: closed for frontier tasks, open for private or high-volume ones.
Hakia 2026
Open hakia.com →From an India-founded ML platform: how gateways make model choice reversible.
TrueFoundry 2026
Open truefoundry.com →The lock-in insurance layer compared hands-on, including India-built Portkey.
Dmytro Klymentiev 2026
Open klymentiev.com →A second, eval-minded view on routing layers so you can switch providers in an afternoon.
Braintrust 2026
Open braintrust.dev →A regularly refreshed plain-English rundown of the current frontier models and their sweet spots.
Codingscape 2026
Open codingscape.com →The most-cited year-in-review; explains why the leader changes constantly and why lock-in is a trap.
Simon Willison Dec 2024
Open simonwillison.net →The independent standard for comparing intelligence, speed, and price on one chart.
Artificial Analysis 2026
Open artificialanalysis.ai →Continuously updated rankings pulled straight from provider APIs; good second opinion.
LLM Stats 2026
Open llm-stats.com →Product-builder-oriented leaderboard that maps benchmark scores to practical use cases.
Vellum 2026
Open vellum.ai →Fastest way to sanity-check what any model actually costs before you wire it in.
Price Per Token 2026
Open pricepertoken.com →The same ground, over in Build the product, our Starting Up track.