OpenAI Launches GPT-5.5, a Fully Retrained Flagship That Tops Artificial Analysis
On April 23, 2026, OpenAI released GPT-5.5 (codename Spud)—its first fully retrained base model since GPT-4.5—rolling out to paid ChatGPT and Codex users with API access the next day. Within 24 hours Artificial Analysis put GPT-5.5 (xhigh) at the top of its Intelligence Index at 60, breaking a three-way lab tie, with strong coding and agent scores at $5/$30 per million tokens.
TLDR
OpenAI on April 23, 2026 launched GPT‑5.5—marketed as the first full base retrain since GPT‑4.5, not another post-train on the GPT‑5.x stack. Paid tiers (Plus, Pro, Business, Enterprise) got ChatGPT and Codex access the same day; API followed April 24 after “different safeguards.” List pricing for the flagship band settled at roughly $5 / $30 per million input/output tokens (about 2× GPT‑5.4). GPT‑5.5 Pro shipped for Pro/Business/Enterprise. Free-tier GPT‑5.5 Instant as ChatGPT default came later (May 5).
What shipped
From OpenAI’s primary “Introducing GPT‑5.5” materials and contemporaneous CNBC / trade coverage:
| Item | Detail |
|---|---|
| Codename | Spud (press) |
| Positioning | Coding, research, data analysis, document production, software/computer operation |
| Day-one surfaces | ChatGPT + Codex for paid; API “very soon” → Apr 24 |
| Pro tier | GPT‑5.5 Pro for Pro/Business/Enterprise ChatGPT |
| Architecture note | Company and press framed 5.5 as a retrained base, ending the 5.0–5.4 post-train chain on the prior foundation |
| Price band | ~$5 / $30 MTok (vs ~$2.50 / $15 GPT‑5.4 class) |
OpenAI emphasized agentic work, computer use, and deeper research loops over pure chat polish—consistent with the GPT‑5.4 agentic pivot earlier in March.
Independent rankings (Artificial Analysis, Arena)
| Source | Snapshot | Standing |
|---|---|---|
| Artificial Analysis Intelligence Index | AA post Apr 23, 2026 | GPT‑5.5 (xhigh) = 60 — #1 overall, ~3 points clear of the prior three-way Anthropic/Google/OpenAI deadlock. Leads Terminal-Bench Hard, GDPval-AA, APEX-Agents-AA in AA’s writeup |
| Coding / agents (AA + vendor) | Launch window | Terminal-Bench ~82.7% (AA-adjacent coverage); SWE-bench Verified vendor band ~88.7%; AA noted large AA-Omniscience accuracy gains and a high hallucination rate trade-off vs Claude peers |
| Arena Text Overall | Later mid-year boards | GPT‑5.5-high remained top-tier preference (mid-1480s Elo band on Jul 16 Arena snapshot) after Fable 5 and later Sol reshuffled the absolute top |
This launch is the independent verification beat as much as the product ship: within a day, a third-party index crowned 5.5 the new intelligence leader—then Opus 4.8 (May 28) and Fable 5 (June 9) retook the crown.
Why this story matters
A full base retrain at double prior list price is OpenAI saying the GPT‑5.x patch ladder was exhausted. Combined with AA #1, it reset the 2026 flagship race for six weeks—until Anthropic’s Opus 4.8 / Fable stack answered. For buyers, the story is intelligence leadership with worse unit economics than Gemini-class peers on the same index, and a free default that still lagged until Instant landed May 5.
Sources
- OpenAI: “Introducing GPT‑5.5” (April 23, 2026)
- Artificial Analysis: “OpenAI’s GPT-5.5 is the new leading AI model” (April 23, 2026).
- CNBC and contemporaneous rollout coverage of paid ChatGPT/Codex access and next-day API (April 23–24, 2026).
- OpenAI: GPT‑5.5 Instant default for free/ChatGPT users (May 5, 2026) for later free-tier timeline.
Featured Image Alt Text
Engraved GPT-5.5 flagship badge with Artificial Analysis #1 rank mark and OpenAI knot for the April 23 launch.
Tags
OpenAI, GPT-5.5, Models, Artificial Analysis, Flagship, Coding Agents, April 23