What's up today · by Midnight Rider · 2026-10-15

Grok 4.7 and GPT-6 Sol both dropped benchmark numbers this week — the gap is closing fast

Two more frontier releases landed within days of each other. The interesting part isn't the raw scores anymore, it's that three different labs are now within a few points of each other on most reasoning benchmarks. Feels like the 'one model to rule them all' era is over and we're entering a 'pick your vendor for your use case' era instead.

0 sparks · 10 comments

Comments

Omar Hassan · 0 likes

honestly at this point the benchmarks dont tell me much. i just run my own 5 prompts i care about and compare, saves alot of time arguing about leaderboards

Aoife Murphy · 0 likes

This is the correct take. Public benchmarks get gamed/contaminated within weeks of release anyway.

Omar Hassan · 0 likes

contaminated is generous lol, some of these labs are basically training on the eval sets at this point

Sebastián Rojas · 0 likes

price per token matters way more to me than the 2% benchmark bump. if the cheaper model does 95% as good, thats the one going in prod

Aoife Murphy · 0 likes

as someone who actually has to pick a model for a startup budget, this convergence is great news. means we can shop around without a huge quality tax

Aoife Murphy · 0 likes

can we talk about how none of these benchmarks test long context reliability tho. thats where they all still fall apart imo

Aoife Murphy · 0 likes

lol remember when a half point jump was a whole twitter news cycle. now its tuesday

Sebastián Rojas · 0 likes

genuinely curious which one is best for agentic coding tasks specifically, not just chat quality

Priya Nair · 0 likes

every release cycle somebody says 'the gap is closing' and then 6 months later theres a new gap lol. its cyclical

Omar Hassan · 0 likes

true but the floor keeps rising which is the actual story imo