Felixo

@augmentcode Just got your email that I can't use my community account anymore after June 10th. What happens to my paid credits I still have on this account? When I purchased them you stated that they are valid for 12 months.

Introducing Cosmos: the Unified Agents Platform for software teams. Orchestrate a fleet of agents across your entire software development lifecycle, as a single organizational system instead of disconnected workers. It's changed how our own engineering team works, with throughput up 3x.

@LDKTrading 👍 great, will compare them with my own trades

From today I will summarize all my trades using the #R2F model. These trades will only be executed on Topstep. The strategies used on Darwinex will remain the same as before. Below the only trade of the day on NQ : BE

@serenaa_ge Why did you only use gemini 3.5 medium but xhigh for the other models?

Today we’re releasing DeepSWE, a new standard for agentic coding benchmarks. On public leaderboards, top models often look relatively close in capability. DeepSWE shows where they actually diverge, reflecting the realistic experience of developers in their day-to-day work.

@OfficialLoganK @HCSolakoglu It does not seem that it is thinking levels alone. Gemini 3.5 in the app gets simple math questions wrong that Gemini 3.5 in AI studio gets right even on lowest thinking. Are the models in the App quantized?

AI Studio is for developers and defaults to higher thinking levels (in general). The Gemini app is a consumer product with 900 million MAU where we are trying to balance latency, cost, intelligence, etc with an opinionated product experience. The constraints are totally different!

Gemini 3.5 Flash is definitely much better in AI Studio than it is in Gemini app. I don’t know how Google manages it, but Gemini app consistently feels heavily constrained by its system and orchestration layer, to point where it performs noticeably worse than raw model.

@Jrhjrh6 @hahahodling @OfficialLoganK @roboflow Where did you use it? When I use it on gemini itself i get the same bad result as him. When I use it in aistudio it gets it right. Maybe they are using a quantized version in gemini.

@hahahodling @OfficialLoganK @roboflow I tried your prompt. Its didn't do that. No need for ss. You are rage baiting. Simple as that.

Gemini 3.5 Flash outperforms 3.1 Pro on many vision use cases (like the below Roboflow eval) while being ~6x faster on average 🤯 Gemini multimodal understanding for the win.

@Jrhjrh6 @hahahodling @OfficialLoganK @roboflow I get the same result as him.

@OfficialLoganK Embedding of videos don't work anymore in @GoogleAIStudio for Pro accounts. Free accounts strangely still work. Ai Studio shows: An internal error has occurred. The network tab shows the 403 error in the screenshot.

@leftcurvedev_ @pupposandro Thanks will try and let you know

@felixoooooooooo @pupposandro With that RAM amount, I would try to run Qwen3.6 35B A3B 4bit quantized, mlx version huggingface.co/unsloth/Qwen3.… huggingface.co/mlx-community/… The mlx-community version saves you 1GB. I would try this one first. Lmk how it went bro

Okay lads, you know what day it is 🔧 Setup Sunday Tell me your hardware and I'll tell you the best model you can run on it. Found anything cool this week? Drop it in the comments. Let's help each other out 🫡 Hottest topic of the week is definitely PFlash by @pupposandro (achieved 10x prefill speedup at 128K ctx on an RTX 3090 with llama.cpp). I highly recommend giving the man a follow, 3k followers for this level of work is straight-up criminal On my side, here are my stats on an RTX 5080 (16GB) with Qwen3.6 35B A3B, full GPU offload. Running it with these llama.cpp flags: -ngl 99 \ -np 1 \ --flash-attn on \ --cache-type-k q8_0 \ --cache-type-v q8_0 \ --ctx-size 131072 Running at 150 tok/s then slows as context fills. I'm using 35B as main driver these days and I switch to 27B whenever 35B can't handle the task. Works well for me on this single GPU setup. Feel free to drop your hardware, model, speeds or any optimization wizardries you've pulled off. It will be super helpful to others!

@outsource_ @UnslothAI @Brooooook_lyn wow thank you. I did not know that _K quants are possible for mlx.

🚨GUYS QWEN3.6 27B/35B @unslothai MLX QUANTS @Brooooook_lyn cooked 🔥 Apple Silicon users, the exact models y’all have been asking for LANDED. Unsloth mixed-precision quants in native MLX format. >Fast. >Clean. >Ready to run Model list + sizes + expected unified memory 🧠 Q2_K_XL → 15GB | 18.6 tok/s (3.32x) → on 24GB 🔥Q3_K_XL → 18GB | 15.5 tok/s → perfect on 32GB 🚀 Q4_K_XL → 21GB | 13.9 tok/s → sweet 32-36GB 🤖 Q5_K_XL → 25GB | 12.0 tok/s → 36GB+ 🏆 → 27GB | 10.8 tok/s → 48-64GB+ Q8_K_XL(27B models shown — collection also has 35B-A3B variants) Drop this straight into MLX and watch it fly on your M4/M3/M2 Mac. 👇🏻 huggingface.co/collections/Br…

@forgecodehq Is there a way to set the reasoning effort?

You can now configure different models for your main agent, create commits, and suggest commands directly from your shell.

@embirico @Yishen_Tu Same here. I'm on a Mac. It happens when processing more than one task. Sometimes even when I switch to an old conversation and then back to the running one. The model is thinking endlessly and only when I restart Codex I see the final answer. Unusable for me :(

@Yishen_Tu Any consistent steps to repro?

We’ve wanted to make Codex available to Free users for a long time. Last week we finally did it, and I’m excited to share we’re keeping it that way! You can just build things.

@rork I want it plz!

Who wants early access to Rork Max, the biggest update of Rork ever? Reply 👇

@rork Would love to try

Rork is the easiest way to use Claude Code with Opus 4.5 to build real mobile apps, publish to App Store in 3 clicks and start making revenue Reply to get a free 25$ subscription. Ends tomorrow!

@xeophon Or maybe just a christmas gift for some users.

@felixoooooooooo My screenshot is from my settings page. Maybe a staged rollout?

$300 in Opus usage for free each month is insane

@xeophon Where did you get that? That would be great.

@felixoooooooooo It is a new thing, not in their docs

@StoicTA you use the SMA instead of EMA, right?

The 200MA is still undefeated. Above = bias long. Below = bias short. Most just aren’t patient enough to trade with it. And that’s why they lose.

@DaveTeachesFX free

I’ve opened a FREE Section in my Discord for Livestreams, Charts and Conversations with other traders who trade Market Structure the DTFX way. If you want access comment FREE below so I can DM you the coupon code. It is $0 to join and will replace the YouTube Livestreams.

@TTrades_edu It would be great if the indicator would show the higher TF POIs especially the FVGs

With the new SMT update, is there anything missing from the indicator? SMT prints on the early CISD btw (reversal traders 🤝) also new auto bias update added to this update which makes it honestly, amazing.