Argus Report
OpenAI's 'Spud' Lands: GPT-6 Base Model Announced After Two-Week Delay, Pitched Directly at Anthropic
The Argus Report OpenAI

OpenAI's 'Spud' Lands: GPT-6 Base Model Announced After Two-Week Delay, Pitched Directly at Anthropic

5 min

After a watched April 14 launch window came and went without a blog post or an Altman tweet, OpenAI today broke the silence: the model internally codenamed Spud is here. Announced April 16 and framed by OpenAI as its “smartest model yet,” Spud is positioned explicitly for “high-value professional work” and designated as a new base system for ChatGPT — the foundation for future GPT-5.5 / GPT-6 iterations. Company President Greg Brockman called it the product of two years of research and a meaningful step toward AGI.

The announcement closes out a volatile two-week window that saw Polymarket contracts, AI-coverage press, and developer Twitter all cycling through increasingly elaborate theories about what had gone wrong.

From March 24 Pre-Training to April 16 Launch

The technical timeline, now clear in retrospect:

  • March 24 — Pre-training completed at the Stargate Abilene, Texas campus on a cluster of 100,000+ H100 GPUs. Altman posted the same day that launch was “a few weeks away.”
  • April 14 — The date most of the market interpreted “a few weeks” to mean. The day passed with nothing shipped.
  • April 14–15 — New consensus window coalesced around April 21 – May 25. Polymarket held 78% probability by April 30 and 95% by June 30. A leaked internal memo — reported by The Decoder — described Spud as making “all OpenAI products significantly better.”
  • April 16 (today) — Official announcement. Post-training, safety evals, and infrastructure readiness — the three outstanding gates — all publicly cleared.

Naming remains deliberately ambiguous. OpenAI has not committed to whether Spud will carry the GPT-5.5 or GPT-6 designation in product UI; the codename “Spud” is being used in the announcement posture itself.

The Pitch: Business, Not Consumer

The most interesting signal in today’s launch is not capability — it is positioning. OpenAI is explicitly reframing around business and enterprise customers. CFO Sarah Friar’s revenue-mix line ran through nearly every press account: business users were 20% of revenue when she was hired in 2024, are now 40%, and are on track to hit 50% by year-end. Spud is the product that’s supposed to pull that number across the line.

This is a direct response to Anthropic. Claude Opus 4.7, which went generally available earlier today, is pitched on exactly the same ground — better coding, better instruction-following, designed to “hand off your hardest work.” Anthropic also paired it with Project Glasswing safeguards as an explicit capability-discipline story versus the unreleased Claude Mythos. Both frontier labs, on the same day, made the same move: a flagship refresh aimed at enterprise procurement desks, with the riskier research frontier held a step behind.

What Resets on the Leaderboard

Before today, the frontier was genuinely contested:

  • Claude Opus 4.6 led SWE-Bench software engineering
  • Gemini 3.1 Pro led GPQA Diamond (graduate-level STEM)
  • Grok 4.20 Beta 2 led medical and legal reasoning
  • GPT-5.4 (released March 5) sat at 83% GDPVal with 1M context — competitive but no longer dominant

Spud rewrites this picture. Benchmark numbers weren’t included in the launch coverage reviewed at publication time, but the competitive framing from multiple outlets (“Mythos-level benchmarks,” per one industry analyst blog) suggests OpenAI is claiming parity-or-better with Anthropic’s internal-only Mythos model. Expect independent evaluations within the week.

The IPO Context Isn’t Going Away

Spud’s timing matters because OpenAI is in active IPO preparation at an $852 billion valuation ($25B annualized revenue). PYMNTS reported earlier this month that backers were questioning the valuation amid the Anthropic competitive challenge. Every week Spud stayed unshipped was a week that narrative compounded.

Today’s launch doesn’t end the IPO-narrative pressure — but it resets it. Pair Spud with the two recent adjacent acquisitions (TBPN on April 6, the media/podcast company; Hiro Finance on April 13, the AI personal-CFO startup) and the IPO roadshow now has the shape OpenAI wants: a flagship model launched, a Codex developer story at 3 million users, an enterprise revenue mix closing in on 50%, and new consumer-adjacent surfaces in media and personal finance.

What To Watch Next

  1. Benchmarks — Independent SWE-Bench, GPQA Diamond, GDPVal, and Terminal-Bench numbers. Expect within 7 days.
  2. API pricing and availability dates — Not disclosed at announcement. This is the enterprise-procurement gate.
  3. Product UI naming — Will Spud ship as “GPT-5.5,” “GPT-6,” or keep the Spud banner? Naming telegraphs positioning.
  4. Anthropic response cadence — Mythos is still unreleased; today’s Opus 4.7 was the pre-emptive move. The next Anthropic ship date is the real competitive read.

The April 14 delay, in retrospect, was a two-day alignment exercise — not a capability slip. OpenAI wanted to ship on the same day as Anthropic’s flagship refresh, not before it. Today, they did.

Key Numbers

  • Pre-training complete: March 24 (Stargate Abilene, 100K+ H100s)
  • Original window: April 14 (missed)
  • Actual launch: April 16 (announced)
  • Polymarket pre-launch: 78% by April 30, 95% by June 30
  • OpenAI valuation: $852B | Revenue: $25B annualized
  • Enterprise revenue mix: 20% (2024) → 40% (today) → 50% target (year-end)
  • Codex users: 3M

Sources: ChatGPT maker OpenAI shifts focus to business users amid Anthropic pressure (Columbian, Apr 16) · OpenAI shifts to business users (Baltimore Sun, Apr 16) · OpenAI’s new ChatGPT base model ‘Spud’: All you need to know (Storyboard18) · OpenAI’s leaked memo says new “Spud” model will make all its products “significantly better” (The Decoder) · OpenAI Spud: Leaked April 16 Release, Mythos-Level Benchmarks (Adam Holter) · OpenAI Release Notes (Releasebot) · OpenAI Backers Question Valuation Amid Anthropic Competition (PYMNTS)