GPT-5 Is Just GPT-4.6—And That’s the Whole Point
A few days ago I was wondering why OpenAI was acting so weird. Now it’s obvious: open-source the old model, fill the U.S. open-source vacuum, bag a $200 M federal contract, stay API-compatible with GPT-5, leak “AGI soon” to juice a $500 B valuation, let employees cash out secondary shares, drop GPT-5… repeat forever.
AGI may not be here yet, but this business choreography is master-class.
Let me say the ugly part out loud: the much-hyped GPT-5 launch is neither a world-shaking revolution nor a clean knockout. If I had to name it, I’d call it GPT-4.6. It’s a cornered “flex,” a spreadsheet move to serve capital before code. AGI can wait—fund-raising can’t.
GPT-5? More like GPT-4.6, a forced flex
Why the downgrade? Because there’s no generational slam. Benchmarks barely nudge past Claude and Gemini by “a point or two”—statistically a rounding error. On coding and multimodal tasks, Anthropic and Google have been wiping the floor with OpenAI for months; even acqui-hires to plug the gap fizzled.
So why ship now?
Simple: had to.
Claude’s enterprise share jumped from 12 % to 32 % in a year; its ARR is rumored to have overtaken OpenAI. The moat is coding—exactly what CTOs pay for. Anthropic’s $1.6 B revenue surge came straight out of OpenAI’s ecosystem (Cursor, GitHub Copilot, etc.). When the competitor is that close, you flip the table: GPT-5’s API price is one-fifteenth of Claude 4.1.
Internally, the launch also ends product anarchy. Research-driven skunkworks spawned a alphabet soup (4.0, 4.5, O3, O4-mini). With researchers poached or burned out, Sam Altmer finally decreed: everything graduates into GPT-5. The number is marketing, not math.
Consumer models plateau; enterprise cash is king
The subtext: consumer LLMs are topping out. trillion-dollar “Stargate” compute clusters aren’t financed by $20-a-month hobbyists. I pay for Plus, but my open-rate is embarrassing. For individuals, good enough is good enough.
The real gale is B2B and B2G. Anthropic CEO Dario Amodei spat the quiet part:
“The best AGI won’t be available to regular people. You’ll get a lobotomized version; the $100 k-a-month clients get the real thing.”
OpenAI’s pivot proves it. GPT-5’s pitch highlights enterprise safety, tool integration, and a simultaneous drop of an open-source cousin, GPT-OSS. Goal: close deals.
Open-source as bait, contracts as trophy: Altman’s four-bird stone
Don’t confuse the repo drop with a return to “Open” roots. The move kills four birds:
1. Win federal checks. White House and DoD deals signed days before the repo went public. Want sovereign contracts? Show source.
2. Occupy the vacuum Meta left. Zuckerberg just said future Llama won’t be open. OpenAI slides in as the U.S. flagship open model, kneecapping Llama’s ecosystem play.
3. Build a funnel. GPT-OSS uses the same toolchain as GPT-5. Start free, hit the ceiling, and the cheapest upgrade path is the paid GPT-5 API. The repo is a lead-gen form disguised as charity.
4. Keep talent. Valuation hype + secondary share sale = liquid millions for staff without burning company cash. Poachers now need a Brink’s truck to compete.
A capital chess game, not a tech leap
Stack the moves: ship a meh-but-cheaper GPT-5 to blunt Claude; open-source to nail government RFPs; pump a $500 B valuation with Stargate theatrics; let employees cash out, locking in talent. Money, market, morale—one loop, zero extra GPUs.
This isn’t a victory for idealism; it’s a finance textbook disguised as a product launch. While rivals obsess over parameter counts, Altman is spinning a flywheel of cash, narrative, and people.
Where’s our opening?
The blitzkrieg reveals a crack: brute scale is running out of miracles. As scaling-law returns diminish, the game tilts from “who burns more” to “who stretches every GPU.” That’s engineering, product polish, cost control—playgrounds where Chinese teams can outrun Silicon Valley budgets.
While U.S. giants wine and dine Capitol Hill, the consumer innovation window is left ajar. Post-training, application-layer, rapid iteration—perfect turf for cheaper, faster, user-obsessed builders.
Forget AGI; learn from Altman: follow the money.
View distilled from my podcast People’s Park Talks AI
👉 Get the AI Daily Brief – two emails a day, one-click listen (VPN required)
Find us:
📺 Bilibili: Call me Xiao Su
📕 Xiaohongshu: People’s Park Talks AI
▶️ YouTube: People’s Park Talks AI