Santh

Never send an organic to do a machine's work.

[Santh @santhreal]

@carlassmann Or you could just use a better harness than OpenCode.

[Santh @santhreal]

@Sharjeelogics Hmm. Have you tried it in a different harness yet, other than agy?

[Santh @santhreal]

Gemini 3.7 is better than half the models on this list.

[Theo - t3.gg @theo]

Rough tier list of where I'd put every major model right now

[Santh @santhreal]

@LexnLin You fell for that? lol

Usage limits are significantly down; I've never even used computer history, images, or any of that.

[Santh @santhreal]

@Sharjeelogics why?

[Santh @santhreal]

@FriesIlover49 @euler2074 @theo @grok also are we going to ignore this?

[Santh @santhreal]

I'll never understand what people who think Gemini 3.7 is worse than Luna and Sonnet use as evals.

[Theo - t3.gg @theo]

Rough tier list of where I'd put every major model right now

[Santh @santhreal]

@FriesIlover49 @euler2074 @theo @grok Yeah, and Gemini 3.7 High is faster than Sol High… Are you missing something, dude?

Also, again. This is still on the faster end of the bench. It doesn't justify putting it below F tier.

[Ben Awad @benawad]

2nd anti-ai sponsor request I’ve gotten

wild to see this

[Santh @santhreal]

@FriesIlover49 @euler2074 @theo @grok @grok did he not just say that it uses 10x more tokens? And is that true or not?

[Santh @santhreal]

dude… did you even read the screenshot? It shows output tokens as well.

@grok read the pictures for him and tell him the difference in output tokens between Sol and Gemini from the screenshots I provided.

And which is faster in terms of e2e latency, considering everything I've shown and whether my definition is correct.

[Santh @santhreal]

@CodeWithOllie @theo the fuck? I provided evidence.

[Santh @santhreal]

Ox Alpha is goated.

No problems with it stopping or not working long enough.

If you guys are having issues... It's a harness problem.

It's worked for about 5 hours straight and is still going(see below)

[Santh @santhreal]

@irinanpic true. Just need to use it with an external harness for it to be good.

[Santh @santhreal]

@blueemi99 crazy cause i dont agree with a single thing on this tier list

[Santh @santhreal]

@xSoloTrades I need help because I said Gemini 3.7 Flash is a good model?

[Santh @santhreal]

watch them spin this like the pen test one:

"Ermm, actually, a human engineer would've taken 1 day and cost more than that. So you actually saved yourself a lot of money."

[Can Vardar @icanvardar]

i ran mythos to merge a pr and it took 30 minutes and cost me $500

[Santh @santhreal]

@Presidentlin lol im going to lose it if I get ratioed by this strawman

[Santh @santhreal]

@euler2074 @BounceNStretch @FriesIlover49 @theo *ttft not ttfs mb

[Santh @santhreal]

E2E Latency is the total number of output tokens, tokens per second, as well as TTFS (Gemini is faster on that too, just not shown below). And it outperforms Sol and the other models in Theo's ranking on most, if not all, of these metrics.

The shown tok/sec for Sol is also very generous, cause Sol is not this fast on the actual sub.

[Santh @santhreal]

@xSoloTrades thanks

[Santh @santhreal]

@PascalLindenau Never said it had to be about just intelligence.

Just thought it was a bad ranking with some egregious choices.