last24hr Deep Dive ← Back
FREE SAMPLE
This is a full deep dive — unlocked so you can see what last24hr subscribers get every day, including the new Polymarket Pulse with three live markets and a divergence note. If you like the format, subscribe for full access to all deep dives, the weekly Quiet Room briefing, and daily Polymarket signal tracking.
Heat 82 — #5 Today AI

OpenAI "Spud" (GPT-5.5 or GPT-6) Pretraining Done — Release Window Opens April 21

FindSkill LumiChats OpenAI OpenAI Index
86%
Polymarket's implied probability that GPT-6 ships by December 31, 2026. 82% by September 30. Pretraining on OpenAI's next frontier model (internal codename: Spud) completed around March 24. Release window per multiple trackers: April 21 – May 25.

The timing is unusually specific. Pretraining finished on March 24, post-training work has been running since, and OpenAI's own release cadence (GPT-5 on July 18, 2025 → GPT-5.2-Codex in mid-April 2026 → GPT-5.4 on March 5) suggests the new model is targeting an April 21 – May 25 release window. Whether it ships as GPT-5.5 or GPT-6 depends on the benchmark delta over GPT-5.4. The naming matters: GPT-6 would be the first post-$122B-raise model and would reset competitive benchmarks within days of Anthropic's Opus 4.7 launch on April 16.

Polymarket is already pricing this conviction. The "GPT-6 released by…?" market has Dec 31, 2026 at 86% and Sept 30 at 82%. That's high — the market doesn't think this is a long shot. Meanwhile, a separate Polymarket market has Anthropic at 93.5% for "best AI model end of April" ($11.77M volume). Those two readings can both be right for a few more days. If OpenAI ships before April 30 and the benchmark gap is meaningful, Anthropic's "best model" lead reprices fast.

Polymarket Pulse — What the Market Is Pricing

Relevant markets

As of Apr 20, 2026
GPT-6 released by…?
Dec 31 2026: 86% Sept 30 2026: 82% Polymarket →

High conviction in the market that the model ships this year. June 30 tier is also active but the specific % wasn't surfaced in today's source data.

GPT-5.5 released on…?
Multi-date buckets Polymarket →

The "Spud gets called 5.5 instead of 6" outcome. Active market with multiple date buckets — the split between this and the GPT-6 market is where the naming bet lives.

Which company has the best AI model end of April?
Anthropic: 93.5% vol: $11.77M Polymarket →

Cross-read for this story. 93.5% Anthropic. Resolves end of April — the same window when OpenAI's Spud might ship. That's the divergence trade.

♦ Divergence to watch Two markets are telling a consistent story right now: Anthropic leads (93.5%) and GPT-6 ships later in the year (86% by Dec). They're consistent because they don't overlap — April's winner vs year-end shipping. But if OpenAI ships this month and beats Opus 4.7 benchmarks, the Anthropic 93.5% number is mispriced. That's where the asymmetric "No Anthropic" bet sits.

The Timeline

March 5, 2026
OpenAI releases GPT-5.4. Incremental improvement over GPT-5.2/5.3. Frontier benchmark position holds.
~March 24, 2026
Pretraining completes on the next major model (codename: Spud). Post-training work begins. Leaker accounts (@apples_jimmy, @btibor91) start posting breadcrumbs.
April 16, 2026
Anthropic ships Opus 4.7. SWE-bench Verified 87.6%. Takes #1 slot on Artificial Analysis Intelligence Index at 57, edging Gemini 3.1 Pro and GPT-5.4.
Mid-April
GPT-5.2-Codex ships. Codex gains 90+ new plugins. Parallel work suggests OpenAI is consolidating its agentic surface area before the Spud release.
April 21 – May 25
Multiple trackers (FindSkill, LumiChats) align on this as the Spud release window. Post-training + safety evals + final benchmarks need to land in the next ~5 weeks.
April 30 (market resolution)
Polymarket's "best AI model end of April" market resolves. If Spud ships before this and beats Opus 4.7 on the index, the 93.5% Anthropic number flips.

Source Divergence: Where the Reporting Splits

Trade press (FindSkill, LumiChats) frames this as a timing question. They have specific release windows (April 21 – May 25) and treat it as imminent. Their coverage reads like a product-launch forecast.

OpenAI's own channels (openai.com/news, openai.com/index) have not confirmed the release date. The GPT-5.2-Codex announcement and Codex plugin expansion are acknowledged; Spud is not. That silence is typical of OpenAI's pre-launch pattern — they don't telegraph frontier releases until days before.

Leaker community on X (@apples_jimmy, @btibor91, plus Sam Altman's cryptic posts) is the noisiest signal. Individually unreliable. Collectively, when multiple leaker accounts start posting within the same week, it's historically correlated with imminent releases — GPT-5 had the same pattern three weeks before launch.

Polymarket is the cleanest read because it's money. 86% by year-end is a hard commitment. The market has been climbing since late March when the pretraining completion rumors started.

The four source types tell one coherent story. Trade press + leakers + Polymarket all point to imminent. OpenAI's silence fits the pre-launch pattern. The convergence is why this story has a heat score of 82 despite being specialty-press-only on the news side — the other three source types are doing the work.

Signal vs Noise

Signal

Pretraining completion + Polymarket at 86% + leaker activity + OpenAI's pre-launch silence pattern. Four independent indicators pointing to a release inside the next 30-45 days. The naming question (GPT-5.5 vs GPT-6) is itself a data point: if it ships as GPT-6, the benchmark delta is big; if it ships as GPT-5.5, it's incremental. Either way the competitive landscape resets versus Opus 4.7.

Noise

Specific release dates from leakers. Leaker accounts often claim specific dates that slip by 1-4 weeks. The window framing is more useful than any single claimed date. Also noise: Sam Altman's cryptic posts. These get dissected but rarely contain actionable information.

Bottom Line

The interesting bet this month is not "does GPT-6 ship" but "does it ship before April 30." Polymarket has two markets in tension: Anthropic at 93.5% for April and GPT-6 at 86% for year-end. If OpenAI executes a late-April surprise release and the benchmark delta is real, the April market repricing is violent. If OpenAI waits until May, Anthropic keeps the month and the two markets remain consistent.

Three things to watch over the next 14 days: a) OpenAI's own blog for any "coming soon" language (their first public signal); b) whether leaker accounts converge on a specific day inside the April 21 – May 25 window; c) any movement in the Polymarket "best AI model end of April" contract. A drop in Anthropic's 93.5% below 85% is the earliest market-side signal that Spud timing is moving forward.

You just read a free sample

Get a deep dive like this every day

  • Daily deep dives on the top 3 stories
  • The Quiet Room weekly briefing (preview)
  • Polymarket Pulse with divergence flags (preview)
  • Full leaderboard access
$25$8/mo
Lock in $8/mo for life
Subscribe
Cancel anytime. 7-day full refund, no questions.

Continue reading the full analysis

Subscribers unlock the complete deep dive plus three extras:

  • All daily deep dives
  • The Quiet Room weekly briefing
  • Polymarket signal tracking (rolling out)
$25$8/mo
Lifetime $8/mo for early subscribers
Subscribe now
Cancel anytime. 7-day full refund, no questions.

Already a subscriber?

Manage subscription →

Disclaimer — last24hr is an information and market-intelligence service. Nothing on this page is financial, investment, legal, tax, or trading advice, nor a recommendation to buy or sell any asset. Content is compiled from public sources and reflects reporting available at time of publication; figures and prices move quickly. Crypto assets and leveraged equities carry significant risk, including total loss of capital. Do your own research and consult a qualified professional before making any financial decision.
← Back