Milhem

Highest CAGR/IQ ratio - Guinness World Records 18, 19, and 22

[Milhem @MilhemGroup]

@jawagner92 It was rough and dirty but yeah, messy 🤷‍♂️

[Milhem @MilhemGroup]

@jawagner92 My honest range is kinda wide C$0.03–0.14/unit, $0.09 base case. Much of the BV burnt up in hotel discount and wind down. Even restricted cash messy.

Too big of a chance of a zero for us little guys.

I didn’t go deep on hotel comps. I’ll probably check when the kids in bed

[Milhem @MilhemGroup]

@perplexity_ai Friendly reminder to always use GLM 5.2 as the orchestrator

[Milhem @MilhemGroup]

@rohanpaul_ai I’m sorry, but in my experience, there’s not a reason to use it

[Milhem @MilhemGroup]

@ChrisUniverse @AskPerplexity @perplexity_ai Sorry to break it. There is a clear value-winner

[Milhem @MilhemGroup]

@jeffgrimes9 GLM 5.2 with no advisor. The secret ticket to good value.

[Milhem @MilhemGroup]

@perplexity_ai Ran the same task through 4 Perplexity Computer orchestrators. One run each, same prompt, same grading rubric.

The cheapest model (GLM 5.2) filed the best report. GPT-5.6 Sol burned 5.9x the credits for a slightly worse one.

Measured in-product, not benchmark-derived.

[Milhem @MilhemGroup]

@therosieum How to fix:

Ran the same task through 4 Perplexity Computer orchestrators. One run each, same prompt, same grading rubric.

The cheapest model (GLM 5.2) filed the best report. GPT-5.6 Sol burned 5.9x the credits for a worse one.

Measured in-product, not benchmark-derived.

[Milhem @MilhemGroup]

@Jason @OpenRouter @bittensor At least the cheapest model is one of the best

[Milhem @MilhemGroup]

@testingcatalog Ran the same task through 4 Perplexity Computer orchestrators. One run each, same prompt, same grading rubric.

The cheapest model (GLM 5.2) filed the best report. GPT-5.6 Sol burned 5.9x the credits for a slightly worse one.

Measured in-product, not benchmark-derived.

[Milhem @MilhemGroup]

@ai_for_success And here’s how you make it affordable

[Milhem @MilhemGroup]

@verge The way to make it affordable

[Milhem @MilhemGroup]

@perplexity_ai Only way to use this affordably

[Milhem @MilhemGroup]

@thesamparr FYI - Ran the same task through 4 Perplexity Computer orchestrators. One run each, same prompt, same grading.

The cheapest model (GLM 5.2) filed the best report. GPT-5.6 Sol burned 5.9x the credits for a slightly worse one.

Measured in-product, not benchmark-derived.

[Milhem @MilhemGroup]

@randomjohnnyh Ran the same task through 4 Perplexity Computer orchestrators. One run each, same prompt, same grading rubric.

The cheapest model (GLM 5.2) filed the best report. GPT-5.6 Sol burned 5.9x the credits for a slightly worse one.

Measured in-product, not benchmark-derived.
I

[Milhem @MilhemGroup]

@AravSrinivas Ran the same task through 4 Perplexity Computer orchestrators. One run each, same prompt, same grading rubric.

The cheapest model (GLM 5.2) filed the best report. GPT-5.6 Sol burned 5.9x the credits for a slightly worse one.

Measured in-product, not benchmark-derived.

[DOMO Capital Management, LLC @DOMOCAPITAL]

Fantastic find @biofuelslaw - this is huge for year-round #E15 in California!!

[Milhem @MilhemGroup]

@AravSrinivas @AnthropicAI What about as the comet browser agent? Swapping out 4.8 would be great.

[Milhem @MilhemGroup]

@perplexity_ai What about as the comet browser agent? Swapping out 4.8 would be great.

[Milhem @MilhemGroup]

@AravSrinivas GLM 5.2 is a game changer for Perplexity Computer equity research.

opus 4.8 is great but credit-efficient prompting isn’t enough given the credit allotment.

GLM 5.2 is 95% quality at 10% the credit cost.

This alone makes a max subscription a no-brainer @AravSrinivas