Jeffrey Forrest

Partner @ Insiders VC. Formerly Lakestar, EA

[Jeffrey Forrest @forrestjeff]

@MikeDenis411611 @WholesomeMeme Just look at the words he uses. “Red cent”. People who post these memes assume billionaires are Scrooge mcduck with a pile of cash. Most times they create valuable IP, jobs, etc - and the world assigns a value to it. Let the builders build.

[Jeffrey Forrest @forrestjeff]

@Sorelle_Arduino Why can’t you just let people be happy.

[Jeffrey Forrest @forrestjeff]

@patrick_oshag @neilmovva Maybe my favorite podcast of yours. Neil is exceptional!

[OrcaRouter 🐳 @OrcaRouter]

A 320B model. On your MacBook Pro. 🐳

We just brought GLM-5.3-Flash natively to Apple Silicon — and invented a new quantization method to make it practical:

OrcaSAQ — Orca Sensitivity-Aware Quantization.

→ 2 / 3 / 4 / 6-bit native MLX
→ Calibration-free
→ Architecture-aware
→ 97.76% Top-1 agreement vs FP8 at 6-bit

320B parameters. Near-FP8 fidelity. A fraction of the memory. Your Mac just got a much bigger model.
https://t.co/ZJiPmypBw7

[Jeffrey Forrest @forrestjeff]

@Austen

[Jeffrey Forrest @forrestjeff]

@FredAlmeidaAMI @cavepaleo 🤦

[Jeffrey Forrest @forrestjeff]

@JordanStev @dannybuck Yeah 4-5k gbp is the better proxy. + he is leaving out asset value and assuming a very high down payment and interest rate.

[Jeffrey Forrest @forrestjeff]

@JeffCrusey @mr_james_c Tis’ an ancient name. A powerful name

[Thibault Jaigu @ThibaultJaigu]

9 models. 12 lead changes. 228 days.

This is the race to be the #1 AI model of 2026, measured by real tokens routed through @RequestyAI.

[Thibault Jaigu @ThibaultJaigu]

Everyone is still quoting model share off OpenRouter.

Open weights are 49.2% of tokens on our gateway and 25.7% of requests. Both numbers are from the same dataset, the same week.

If you quote share of requests you get one market. Share of tokens you get a different one. Share of spend, a third. Pick the one that matches the decision you are making, and say which one you picked.

[Jeffrey Forrest @forrestjeff]

@signulll @hkanji This is horrifying - day ruined lol

[Jeffrey Forrest @forrestjeff]

@imbabybrooklyn @NousResearch This is too cool. Can't wait to use 🔥

[Jeffrey Forrest @forrestjeff]

@paulbonnet @SaidHaschemi This is insane 🤯

[Jeffrey Forrest @forrestjeff]

@oelma__ Tennisball

[Jeffrey Forrest @forrestjeff]

@modic123 @OpenAI Interested in the full case study if you're sharing it!

[Chirag Modi @modic123]

NextSlide joining @OpenAI is a decent signal that the labs are accelerating their push into office workflows. After all, slides and spreadsheets kind of rule the corporate world.

Great slides have to do several things at once: get the analysis right, tell a coherent story, keep the numbers consistent and look polished enough to use. None of these is optional in a real workflow.

As a fun little experiment, we gave four LLMs the same assignment: build a Private Equity Investment Committee presentation (25+ slides) on a potential take-private of EPAM Systems, a ~$5bn IT-services company.

An expert (ex-PE professional) rated the AI output decks (slide by slide) on various dimensions including: (1) Structure and Completeness, (2) Logic and Narrative, (3) Numerical Accuracy, (4) Internal Consistency, (5) Formatting and Visual Polish.

Those slide-level judgments were aggregated into an overall score for each deck. The results were pretty interesting:

- Opus 5 Extra: 51/100
- GPT-5.6 Sol xHigh: 42/100
- GPT-5.6 Terra xHigh: 30/100
- GPT-5.6 Luna xHigh: 25/100

Formatting and visual polish was the weakest area across the board.

Most of the decks looked impressive at first glance. None was close to something an investment team could use as-is.

My takeaway as a former banker: there is still quite a bit of headroom, but the direction is very promising.

Thinking of scaling this into a public benchmark. Happy to share the full case study - DMs open!

[Jeffrey Forrest @forrestjeff]

@HarleyInHD @michael_chomsky Hermes desktop a lot more approachable for the masses imo

[Jeffrey Forrest @forrestjeff]

@witcheer @michael_chomsky Awesome move to Nous - congrats :-)

[Jeffrey Forrest @forrestjeff]

Important point from @DavidSacks from the last All-In on Leopold Aschenbrenner’s fund:

Going from $225M to $45B in AUM does not mean a 200x return. It’s an open-ended fund, so new capital can subscribe at any time.

The fund was up ~450% YTD and is now up ~80%. If most of the capital arrived after the early gains, most investors never earned anything close to 450%.

AUM growth ≠ investment returns.

[Jeffrey Forrest @forrestjeff]

@PimDeWitte Insane stuff 🚀🚀