Here's a story that reads like a soap opera script. On September 12, Anthropic CEO Dario Amodei published an essay urging the industry to curb rapid capability jumps in AI. Within hours, Elon Musk said "Dario is right," Sam Altman agreed, and Google DeepMind's Demis Hassabis called it "the right path forward." A genuine AI slowdown consensus — for about 72 hours. Then both Altman and Musk went right back to teasing their next powerful models, and honestly, nobody who follows this space was surprised.
01.The slowdown agreement that lasted a weekend
Amodei's essay landed on September 12 and became the most-discussed AI document of the month. "We must slow the space in which we improve the capabilities of AI models," he wrote, calling for industry-wide coordination to pace the frontier. As we covered when the antitrust lawsuit dropped, that public show of unity has already landed all four companies in a US federal court — plaintiffs allege the mutual agreement to slow development violates antitrust law.
So the timing here matters. On September 15 — three days after agreeing that everyone should take a breath — Altman replied to a post about shipping pace with: "Big ship this week and then for devday." He'd earlier hinted that this week's release volume would be on par with what you'd expect from DevDay 2025. Not exactly a company pumping the brakes.
02.Musk's Grok 4.8: 2.5 trillion parameters
Musk, meanwhile, laid out a roadmap on September 14 that is the opposite of slowing down. SpaceXAI's upcoming Grok 4.8 — with a massive 2.5 trillion parameters — is set to complete pre-training and move into reinforcement learning this week. In his own words: "Grok 4.7 should be roughly on par with Opus 5.0, not 5.1. Better in some ways, worse in others. We need to fix multimodal performance. Grok 4.8 will be a noticeable improvement. Grok 4.9 is probably Astra/Fable class. Grok 5 maybe better than anything. We shall see."
For context on how close this race actually is: SpaceXAI claims Grok 4.6 scored 61 on the Artificial Analyst Intelligence Index, ahead of OpenAI's GPT 5.6 Sol and a single point behind Anthropic's Fable 5. If you enjoy tracking this stuff, we broke down Musk's earlier Grok roadmap here — the pace since then has only accelerated.
03.The pricing war is the part Indian builders should watch
Here's the thing nobody's talking about enough: the price gap. SpaceXAI's Grok 4.6 is billed at $2 per million input tokens and $6 per million output tokens. Anthropic's Fable 5? $10 and $50 for the same. That's one-fifth the input cost and roughly one-eighth the output cost. For Indian students and freelancers building on APIs — where every dollar of API spend converts directly to rupees out of your pocket — this kind of undercutting is the real news. Aggressive pricing is how a challenger eats a market, and it forces everyone to respond.
If you're building on these APIs, learning to structure your prompts and cache what you can is no longer optional — it's the difference between a hobby project and a business that survives its first invoice. The cost-optimization chapter in our GPT-6 Astra — Zero to Hero Guide was written precisely for moments like this, when prices start moving fast.
04."Our C/C++ training stack was written by humans"
The week's spiciest side plot: a Google AI researcher suggested SpaceXAI's internal architecture likely leans on code generated by rival models like GPT or Fable. Musk pushed back hard, insisting the underlying C and C++ training stack was written entirely by human engineers — though he didn't clarify whether his teams used AI coding agents alongside them. His line: "AI is not yet good enough for extremely high-performance software (but it will be)." Make of that what you will; the pettiness between these labs is now a permanent feature of the news cycle.
05.Why this matters beyond the drama
Strip away the personalities and two things remain. First, every major lab is simultaneously calling for caution and sprinting at full speed — a contradiction that courts, regulators and lawmakers are now actively probing. Second, the competitive pressure is real: if Grok 4.8 delivers a "noticeable improvement" as promised, OpenAI and Anthropic will have to answer, slowdown talk or not. That means better, cheaper models arriving faster — which, for users in India, is quietly excellent news.
Key Takeaways
- Altman and Musk publicly backed Amodei's September 12 AI slowdown call, then teased aggressive releases within days
- Grok 4.8 (2.5 trillion parameters) finishes pre-training this week; Grok 4.9 is targeted at "Astra/Fable class"
- OpenAI has a "big ship" coming this week plus a DevDay-scale release wave after
- Grok 4.6 undercuts Fable 5 by 5x on input tokens and ~8x on output tokens
- The pacing "consensus" is now the subject of a federal antitrust lawsuit
Frequently Asked Questions
What did Sam Altman and Elon Musk announce after the AI slowdown call?
Days after backing Dario Amodei's call to slow frontier AI, Sam Altman promised a 'big ship' this week plus a DevDay-sized wave of releases, and Elon Musk said Grok 4.8 — a 2.5-trillion-parameter model — is finishing pre-training and moving into reinforcement learning, with Grok 4.9 targeted at 'Astra/Fable class'.
How much cheaper is Grok 4.6 than Anthropic's Fable 5?
Grok 4.6 is billed at $2 per million input tokens and $6 per million output tokens, versus $10 and $50 for Anthropic's Fable 5 — one-fifth the input cost and roughly one-eighth the output cost. For Indian students and freelancers paying API bills in rupees, that price gap matters more than any benchmark score.
Is the AI slowdown consensus still holding?
On paper, yes — after Amodei's September 12 essay, Altman, Musk and Google DeepMind's Demis Hassabis all endorsed pacing the frontier. But that public show of unity is now the subject of a US federal antitrust lawsuit alleging the labs coordinated a slowdown, and both Altman and Musk went right back to teasing aggressive new model releases within days.
What This Means for You
If you're a developer or student in India watching this from the sidelines, don't get distracted by the CEO drama — focus on what it delivers: falling prices and improving models, month after month. Every pricing cut like Grok's makes side projects more viable and AI-powered services more profitable. The winners won't be the people arguing about which CEO broke his promise; they'll be the ones who've already learned to direct these models well. If you want to be in that second group, our GPT-6 Astra — Zero to Hero Guide takes you from complete beginner to building working AI agents — including the cost-optimization skills that matter most when prices start falling this fast.
Confused by all the GPT-6 Astra buzz lately? We wrote a plain-English beginner's guide to GPT-6 Astra that is worth a read before you spend a single rupee on AI tools.