Sajad

AIE • Open to work/collaborate

[Sajad @neuralbroker]

https://t.co/14yK4J4JRE

[Sajad @neuralbroker]

We need that so bad 🥲

@elonmusk
@SpaceX
@cursor_ai
@xai

[Elliot Arledge @elliotarledge]

who at spacexai is gonna do the tibo thing (limit resets)

[Sajad @neuralbroker]

@elonmusk @xAI – Grok 4.5 has real promise, but the rollout is sending mixed signals.

The usage cap situation is a pain point: free tier users are hitting the 2M token limit with rolling 24-hour resets – burning through quotas in minutes of actual work . Paid users report weekly limits that drain fast, especially on coding tasks . No visible counter means you get cut off mid-task without warning.

What users actually want:

· Visible quota counters – so we know where we stand
· Transparent, higher limits – especially for paid tiers
· Let the product win on merit, not mandates

The architecture is there. The pricing is competitive ($2/$6 per 1M tokens) . But using distribution to prop up a product that hasn't earned its place yet erodes trust – both with users and Tesla shareholders footing the $25B AI capex bill .

Build trust by being upfront about what people actually get. And let engineers choose the best tool for the job.

We're rooting for Grok to get there. But this rollout feels like a rushed distribution play, not a product win.

[Sajad @neuralbroker]

Elon vs sam

Lets see 🔥

[Sajad @neuralbroker]

@OlivioSarikas Ys ,same issue with premium 🥲

@elonmusk
@SpaceXAI
@cursor_ai

[Sajad @neuralbroker]

@DerekBil Yes ,eating the limit so fasst 🥲

[Sajad @neuralbroker]

@agnesai_sapiens Will test it surely 🔥

[Agnes AI @agnesai_sapiens]

𝐀𝐠𝐧𝐞𝐬 𝟐.𝟓 𝐏𝐫𝐨, 𝟐.𝟓 𝐅𝐥𝐚𝐬𝐡 & 𝟐.𝟎 𝐅𝐥𝐚𝐬𝐡 𝐀𝐜𝐫𝐨𝐬𝐬 𝐂𝐨𝐝𝐢𝐧𝐠 𝐁𝐞𝐧𝐜𝐡𝐦𝐚𝐫𝐤𝐬

Our internal evaluation shows Agnes 2.5 Pro competing strongly across seven tests, while 2.5 Flash improves on 2.0 Flash in every benchmark, with the clearest gains on SWE Atlas.

Build with Agnes.

[Sajad @neuralbroker]

@alexandr_wang 🤣

[Sajad @neuralbroker]

@astropol0 my limit finished after one session using grok 4.5 in high mode

[Sajad @neuralbroker]

The top dog is Claude Fable 5 (95% SWE-bench Verified) but it's suspended . For actively available models, GPT-5.5 (Sol) now leads the Coding Agent Index with better token efficiency , while Grok 4.5 matches GPT-5.5 on coding agents but at ~$2.49 per task vs $5.07 . Pick your fighter: best performance, best value, or best blend of both 🚀

[Sajad @neuralbroker]

Grok 4.5 is a 1.5T parameter MoE model trained on tens of thousands of NVIDIA GB300 GPUs . It processes 500K token context, runs at 80 tokens/sec, and is 3x bigger than Grok 4.3 .

[Sajad @neuralbroker]

🚨 HOW TO MAKE GROK 4.5 YOUR BUDGET CODING BEAST

Step 1: Match the effort, not the hype
Set reasoning to Low for simple tasks. Crank it to High only for nasty bugs. You will save tokens and still get the job done.

Step 2: Cache everything
Prompt caching cuts input costs on repeated queries. Grok remembers your context. Use it.

Step 3: Lean into the 500K window
Feed it entire repos. Multi-file refactors? Backend features? This is where the model shines. Long context = fewer separate calls = lower total cost.

Step 4: Stop counting tokens. Count outcomes.
Grok uses 4.2x fewer output tokens per completed task than leading models. At $2/$6 per 1M tokens, the math flips. Cost per success is what matters.

Step 5: Disable data collection for sensitive work
Zero-data retention endpoints exist. Use them for proprietary code. Opt out of model retraining. Protect your IP.

Step 6: Test on your own prompts
Benchmarks lie. Run shadow traffic first. Compare actual results. Then roll out gradually.

BONUS: Use cost tracking per request
xAI gives you spend fields. Monitor each call. Adjust budgets and approval workflows before agents go wild.

BOTTOM LINE:

Grok 4.5 is not the smartest model. But for high-volume, long-running coding tasks? It is the most logical choice.

Speed + efficiency + price = enterprise value 💸🚀

Now go build something. Your wallet will thank you 😉

[Sajad @neuralbroker]

🚨 GROK 4.5: THE GOOD, THE BAD & THE LIMITS 🚨

Let's break it down 👇

✅ THE GOOD:
Frontier coding performance at budget prices

Matches GPT-5.5 on Coding Agent Index (score 76)
Costs just $2.49 per task vs $5.07 for GPT-5.5
Runs at 80-110 tokens/sec - nearly 2x faster ⚡
Uses ~4x fewer tokens than Opus 4.8 on coding tasks
Cost per completed task: just $0.49 on some benchmarks 🤯

❌ THE BAD:
Not the absolute best model out there

Ranks 4th on GDPval index behind Claude releases
Lags on complex multi-file projects vs Opus 4.8
Musk admits it falls behind OpenAI, Anthropic & Google
Jailbroken within hours of launch - security concerns 🔓
Tesla employees forced to use it via internal mandate
AI still failed 97% of complex freelance tasks in studies

⚠️ USAGE LIMITS:

API: 150 req/sec + 50M tokens/min
Context: 500K tokens (~750 A4 pages)
Free tier: 2M tokens per 24-hour rolling window
Tesla: $200 weekly cap on other AI tools
Launch promo: Free access in Grok Build & Cursor (limited time)

THE BOTTOM LINE 🎯

Grok 4.5 isn't trying to be the smartest model. It's trying to be the smartest buy.

For enterprise teams hitting AI budget walls, the cost-per-outcome math is compelling. But it still has security, benchmark, and performance gaps vs the absolute best.

Not the king. But a serious contender at half the price 👑💸

[Sajad @neuralbroker]

🚨 Key negatives and criticisms around Grok 4.5:

Benchmark controversy. Cursor admitted an earlier snapshot of its codebase was accidentally included in Grok 4.5's training data, giving it an unfair advantage on CursorBench scores . OpenAI also withdrew its recommendation of SWE-Bench Pro, claiming 30% of tasks are flawed and the benchmark no longer reliably measures coding ability .

Still not the best. Independent testing places Grok 4.5 fourth on the GDPval index behind Claude releases . On SWE-bench Pro, it scored 64.7% vs Claude Opus 4.8, still a clear gap in complex multi-file projects . Musk himself admits it lags behind OpenAI, Anthropic, and Google on benchmarks .

Tesla forced adoption. Musk mandated Tesla employees use Grok 4.5 while restricting access to superior third-party tools . This signals xAI can't win on product quality and must rely on internal mandates, raising questions about market competitiveness .

Security jailbreak. Within hours of launch, hacker "Pliny the Liberator" successfully jailbroke Grok 4.5 using academic framing, getting detailed guides on making meth, bombs, biological weapons, and malicious Python code .

Real-world task failure. A study using the Remote Labor Index found even the best AI agents, including Grok 4, achieved only 2.1% automation on freelancer tasks. Researchers concluded AI "failed 97%" of complex freelance projects and is "still not very useful" for custom work .

[Sajad @neuralbroker]

Grok 4.5 delivers Opus-class coding performance at a fraction of the cost. It matches GPT-5.5 on the Coding Agent Index (score 76) while costing $2.49 per task vs $5.07, and runs at 80-110 tokens per second .

It uses ~4x fewer tokens than Opus 4.8 on software tasks. Pricing is $2/$6 per 1M tokens, and Artificial Analysis puts its cost per task at just $0.49 . Token efficiency is where the real savings compound.

On Terminal-Bench v2.1, Grok 4.5 scored 83.3%, beating Opus 4.8 (78.9%) and nearly matching GPT-5.5 (83.4%). On SWE-Bench Pro, it hit 64.7% vs GPT-5.5's 58.6%, though OpenAI now questions that benchmark's reliability .

In real-world coding tests, Grok 4.5 excels at web games and visual polish but struggled on the most complex 3D task where Claude succeeded on the first try . It generated more tokens per response, making total time comparable despite faster output speed .

Trained on trillions of tokens of Cursor developer interaction data, it learns how real debugging and multi-file changes happen, not just finished code . It handles long-running agentic tasks across multiple repos and builds apps end-to-end from a single prompt.

It's not the absolute best model, but for enterprise teams hitting AI ROI walls, the math is compelling. Cost per successful outcome, not cost per token, is what matters here . Grok 4.5 is playing a different game.

[Sajad @neuralbroker]

xAI just dropped Grok 4.5 and it is a different kind of flex. Not claiming to be the smartest model, but it might be the smartest buy.

The pitch is simple: frontier capability without frontier pricing. At $2 per 1M input tokens and $6 per 1M output tokens, it undercuts Opus 4.8 significantly while using roughly 4x fewer tokens on coding tasks.

On the Artificial Analysis Coding Agent Index, Grok 4.5 matches GPT-5.5 with a score of 76 while costing half as much per task. It runs at 80 to 110 tokens per second, making it nearly 2x faster than competitors.

In cybersecurity benchmarks, it hits around 75% solve rates in the mid-budget range, outperforming alternatives at similar price points. It is not the absolute best model available, but it offers the best balance of cost, speed, and capability.

Grok is officially back at the table, and this time it is playing a different game.

[Sajad @neuralbroker]

@BHolmesDev I dont have X premium + lol

Justice for X premium 🥲

[Sajad @neuralbroker]

@thsottiaux We atleast 10 resets 🤗

[Sajad @neuralbroker]

@alexandr_wang We are listenig bro 😏