Opinions expressed by Entrepreneur contributors are their own.
Key Takeaways
- Nvidia CEO Jensen Huang declared that “AGI has arrived” following OpenAI’s GPT-6 Astra launch, but OpenAI itself never made that claim, and researchers call the term “jagged” fairly than general.
- Astra and Claude Fable 5.1 article near identical scores on multiple benchmarks, yet create noticeably distinct genuine earth output. The gap traces rear to the “harness” wrapped about the model, not the example itself.
- SpaceX’s $60 milliard acquisition of Cursor shows anywhere the genuine rivalrous combat is moving. It’s not the raw example anymore. It’s who controls the tier about it.
AGI, artificial broad intelligence, method a scheme that can equivalent or attack individual achievement throughout basically any cognitive task, not fair the narrow ones it was trained for. That difference matters since it’s exactly what’s being asserted concerning correct now.
Jensen Huang has a habit of declaring triumph at the exact instant his business needs the market to accept it. So whenever OpenAI shipped GPT-6 Astra on Sept. 3, Huang posted on X that “AGI has arrived,” adding that “400K GPUs” are “coming online next.”
I’m an AI engineer, and I’m currently construction ChatMakr, an AI-powered phase that creates multi-agent chatbots in seconds. I spent the final twelvemonth observing all example publish get relabeled as the one that eventually changes everything. Here is the truth.
What Astra really does
Astra isn’t a chatbot upgrade. OpenAI positions it as an controller — a scheme that browses, codes, runs multi-step workflows, models in Blender and handles cybersecurity tasks alongside minimal supervision. The benchmark numbers are genuinely striking: 97.6% on FrontierMath Tier 4, 99.9% on ARC-AGI-3, and a ideal 100% on ExploitBench, up from 78.5% for its predecessor, GPT-5.6 Sol. Those are not incremental gains.
But OpenAI itself, notably, stopped abbreviated of calling Astra AGI, choosing alternatively to emphasis “speed and versatility.” That restraint from the business that built the example is the tell.
Why group aren’t convinced
The pushback isn’t really concerning whether Astra is impressive. It plainly is. It’s concerning the term AGI itself. Wharton prof Ethan Mollick has asserted for months that what we’re seeing is “jagged AGI:” systems that are superhuman in several narrow domains and unreliable in others, which is a extremely distinct item from broad intelligence.
There’s motionless no accepted upon test for AGI, so declaring it “arrived” is nearer to a subjective judgement call than a measurement. Augmented Mind and Manolo Remiddi put it plainly: “AGI is the item anywhere you can rely the output without checking. We are not there.”
The Fable 5.1 difference and why scores lie
This is anywhere it gets engaging for anyone really construction alongside these models. On paper, Astra and Anthropic’s Claude Fable 5.1 appearance neck and neck on multiple coding and reasoning benchmarks. In practice, engineers study meaningfully distinct output quality. Fable tends to create additional production-ready code alongside small output tokens, during Astra’s power shows up additional in device use and multi-step automation. Snorkel AI’s testing revealed that Fable fails through a basically distinct form than how it succeeds, losing valid activity mid-trajectory fairly than operating out of budget, which matters far additional to a shipping squad than a leaderboard percentage.
The logic akin scores can conceal specified distinct outputs comes downward to item most group exterior AI engineering have never heard of: the harness.
What a “harness” really is
A harness is the infrastructure wrapped about a model: the immediate assembly, tool access, recollection administration and error improvement that decide how the raw example is really put to work. As MindStudio describes it: “A example is the engine. The harness is the car. Most benchmarks test the car.” Swap the harness and the identical example produces wildly distinct results. Cursor’s own investigation established one example scoring 46% accuracy alongside one harness and 80% alongside another, alongside no example alter at all.
Astra’s own benchmark former makes the item improved than any explainer could. On ARC-AGI-3, Astra scored 62.7% using a norm harness, basically tied alongside its predecessor, and 99.9% using a “Provider Adapter” harness that let it keep and reuse previous reasoning alternatively of restarting chilly on all step, according to Augmented Mind and Manolo Remiddi. Same weights. Same model. A 37 item sway from infrastructure alone. That’s not a small footnote. It method a huge portion of what gets reported as “model improvement” is really harness improvement, and leaderboards that don’t authority for it are telling a story, not reporting a fact.
Why Cursor fair became the most crucial acquisition of the year
This is exactly why SpaceX’s $60 milliard acquisition of Cursor matters additional than it looks. Cursor isn’t a example company. It’s among the finest harness companies in the world, the tier that decides how coding models really get used. Folding it into SpaceX’s GPU fleet is a bet that owning the tier is as precious as owning the model, perchance more. OpenAI has already cut Cursor’s admission to its models complete rely concerns tied to Musk’s companies, which tells you the example providers see the identical thing: Whoever controls the harness controls the merchandise experience, despite of which example sits underneath it.
The base line
Nobody grave disputes that GPT-6 Astra is a important model. What’s contested is the jump from “major model” to “AGI has arrived.” If you’re evaluating these systems for your own business, disregard the AGI framing entirely. Look at the harness a vendor wraps about the model, ask how it fails, and test it on your genuine workflow. The benchmark gap between Astra and Fable 5.1 already proved the leaderboard solitary won’t inform you.
Key Takeaways
- Nvidia CEO Jensen Huang declared that “AGI has arrived” following OpenAI’s GPT-6 Astra launch, but OpenAI itself never made that claim, and researchers call the term “jagged” fairly than general.
- Astra and Claude Fable 5.1 article near identical scores on multiple benchmarks, yet create noticeably distinct genuine earth output. The gap traces rear to the “harness” wrapped about the model, not the example itself.
- SpaceX’s $60 milliard acquisition of Cursor shows anywhere the genuine rivalrous combat is moving. It’s not the raw example anymore. It’s who controls the tier about it.
AGI, artificial broad intelligence, method a scheme that can equivalent or attack individual achievement throughout basically any cognitive task, not fair the narrow ones it was trained for. That difference matters since it’s exactly what’s being asserted concerning correct now.
Jensen Huang has a habit of declaring triumph at the exact instant his business needs the market to accept it. So whenever OpenAI shipped GPT-6 Astra on Sept. 3, Huang posted on X that “AGI has arrived,” adding that “400K GPUs” are “coming online next.”
I’m an AI engineer, and I’m currently construction ChatMakr, an AI-powered phase that creates multi-agent chatbots in seconds. I spent the final twelvemonth observing all example publish get relabeled as the one that eventually changes everything. Here is the truth.
