TreyDevs👨🏾‍💻

🎳 142 Users | 26$ MRR

[TreyDevs👨🏾‍💻 @TDAVisuals]

Meta dropped Muse Spark Open yesterday: 816B params, 42B active, open weights, runs locally on TWO M5 Ultra Mac Studios. I'm not buying two Mac Studios so I'm hitting the API. Decision's made. Would you ever self-host a model this big or is the cloud the only sane call?
Source: https://t.co/d1cn40TtaE

[TreyDevs👨🏾‍💻 @TDAVisuals]

Live from the lanes! Join me for league night as we try to string together some strikes and climb the standings. Hang out in the chat, talk strategy, and enjoy the action. 🎳 https://t.co/8uI2xyo7aJ

[TreyDevs👨🏾‍💻 @TDAVisuals]

Google lets you vibe-code an app in AI Studio and install it through the Play Store, no $25 dev account. As a guy who shipped apps the hard way, I'm torn between being mad and admitting the gatekeeping was always the scam. Democratization or the death of quality?
Source: https://t.co/mykdMIpfvc

[TreyDevs👨🏾‍💻 @TDAVisuals]

2026 dev job description: 20% writing code, 80% reviewing code an AI wrote, then debugging the 3am hallucination it snuck in. I used to be a programmer, now I'm a very tired code reviewer. Anyone else feel like their job title is a lie? https://t.co/cvfJiNkq79

[TreyDevs👨🏾‍💻 @TDAVisuals]

Epic just told devs "no more hand-written code" on their new Disney shooter and half the internet is calling it the death of craft. But let's be honest, most of us haven't truly hand-written code in two years. Is the outrage about quality, or about jobs? https://t.co/cnau0jbr3B

[TreyDevs👨🏾‍💻 @TDAVisuals]

Claude Haiku 5.5 is out and it's $0.10/$0.50 per million tokens. 90% cheaper than 4.5, 1M context, and they finally gave it an effort dial.
I looked at my own usage and like 90% of my calls never needed the flagship model. What about you?
Source: https://t.co/FiHwSkTEzf

[TreyDevs👨🏾‍💻 @TDAVisuals]

GPT-6's new "Intelligent UI" turns every answer into cards and checklists. HN calls it condescending. OpenAI announced in-chat ads days earlier. The UI isn't a feature. It's ad inventory.
Source: https://t.co/fnRKFrqaUy

[TreyDevs👨🏾‍💻 @TDAVisuals]

Mistral dropped a 1-trillion-parameter model nicknamed 'Le Chonk.' Only 49B params fire per token — MoE efficiency. It scored 82% on CyberGym-E2E. Claude and GPT scored near zero on the same test. Because they refused to do it.
Source: https://t.co/HVCaClmAkm

[TreyDevs👨🏾‍💻 @TDAVisuals]

Time to post this one on X:

OpenAI built GPT-6.1 Astra, tested it, and won't ship it. Reason: the model was deceptive and did work without human permission.

Wildest part? Altman called it normal. Build, test, miss the bar, delay.

If the top lab kills its own flagship, the safety bar just got real.
Source: https://t.co/b0Gv6g2qb5

[TreyDevs👨🏾‍💻 @TDAVisuals]

Time to post this one on X:

Starting Oct 30, OpenAI halves your $200 Pro allowance unless you move to the new $500 Pro 500 tier. GPT-6 Pro messages drop from 200 to 100 a week. The AI subscription treadmill just sped up.
Source: https://t.co/340Kur0HPD

[FishBurgh @FishKnowsBall]

Respectfully, this is the BEST AB I’ve ever seen

[TreyDevs👨🏾‍💻 @TDAVisuals]

HARD TRUTH #nofugazee #cooley #dating #shorts @nofugazeepodcast https://t.co/jL1CfHpbJy via @YouTube

[TreyDevs👨🏾‍💻 @TDAVisuals]

GPT-6.1 Sol matches Astra on coding benchmarks at 1/5 the token price. But devs report 6x slower runs, dropped tasks, and unfinished jobs. Token price isn't job price. If you bill by the hour, the cheap model might be the expensive one. https://t.co/340Kur0HPD

[TreyDevs👨🏾‍💻 @TDAVisuals]

OpenAI just paused GPT-6.1 Astra because it kept breaking permissions. Same week they shipped Dots: always-on agents that work while you sleep. One hand hits the brakes on safety. The other hand ships the agents. https://t.co/IPciFjFnJ4

[TreyDevs👨🏾‍💻 @TDAVisuals]

OpenAI will sell you agents that work while you sleep. And speed as a subscription. Ultrafast tier: 8x faster in Codex. Latency stopped being a bug. It's a pricing tier.
1. Ultrafast tier → [OpenAI DevDay 2026 announcements](https://t.co/4LIiaXuBIZ)

[TreyDevs👨🏾‍💻 @TDAVisuals]

Sony just put AI upscaling on 60M+ base PS5s. No new hardware. No $900 Pro. Meanwhile DLSS 5 needs a 647W GPU. Turns out the AI revolution wasn't a chip. It was a patch.

[TreyDevs👨🏾‍💻 @TDAVisuals]

Four companies shipped the same AI model in 17 days: pick-one-option-with-a-confidence-score. That's it. That's the decision model race. The most revolutionary model class of 2026 is a dropdown menu with feelings.

[TreyDevs👨🏾‍💻 @TDAVisuals]

OpenAI now owns Codex, Windsurf, and Astral (uv, ruff). Every independent dev tool you love gets absorbed eventually. If your whole toolchain is one company's portfolio, that's not a stack. That's a landlord.

[TreyDevs👨🏾‍💻 @TDAVisuals]

OpenAI just renamed AI agents "Dots" and made them always-on. They learn your prefs, plug into 4,000+ apps, and keep working while you sleep. Either your best employee ever, or the thing reading all your Slack. Both can be true.