GPT 5.5

OpenAI's flagship model released April 2026 (codenamed "Spud"). Positioned as "more with less" — fewer output tokens per task, faster autonomous decomposition of vague prompts.

Key facts

  • Type: Foundation language model
  • Maker: OpenAI
  • Released: April 2026
  • Codename: Spud
  • Pricing: $5/M input tokens, $30/M output tokens (double GPT 5.4; marginally more expensive than Opus 4.7 on output)
  • Context window in Codex: 400K tokens (vs Opus 4.7's 1M)
  • Availability: ChatGPT, Codex; direct API coming soon at launch
  • Codex role: One of the selectable Codex models in Nate's May 2026 Codex app walkthrough, alongside GPT 5.4 and other model choices [src-048]
  • Reasoning controls: Codex exposes speed/intelligence settings such as low, medium, high, and extra-high reasoning in the app interface [src-048]
  • Roberts's framing: GPT 5.5 is presented as the strongest Codex model option and paired with a speed/performance toggle that trades usage for faster execution [src-058].

Benchmark comparison vs Claude Opus 4.7 (Nate's 4-task coding benchmark)

Metric GPT 5.5 Opus 4.7
Total runtime (4 tasks) 20m 49s 40m 43s
Output tokens ~70K ~250K
Relative cost ~$3 cheaper baseline
Terminal Bench 2.0 82.7 69.4
SWE-bench Verified lower higher

Key finding: GPT 5.5 uses ~3.5x fewer output tokens for equivalent outputs — the "more with less" positioning is measurable. [src-012]

Related

Source references

  • [src-012] Nate Herk — Video editing & content creation cluster (2026-04-15 to 2026-04-23)
  • [src-048] Nate Herk — "Master 97% of Codex in 1 Hour (full course)" (2026-05-06)
  • [src-058] Jack Roberts — "How to use Codex Better than 99% of People" (2026-05-06)

Robin Cartier perspective

This page is part of Robin Cartier's working AI knowledge graph: a practical research layer for production AI, recommendation systems, experimentation, GEO, and agentic web readiness.

The useful next step is to connect this concept back to applied product leadership and operating models.

Recommended next

Keep reading from this thread

From 500 indexed pages and articles.

  1. Wiki concept Practitioner Model Benchmarking Methodology Nate Herk's framework for evaluating AI models with real agentic tasks, beyond official benchmarks. Related by output
  2. Wiki concept Codex Chronicle Screen Memory The research-preview Codex memory layer that captures recent screen context, using screenshots/OCR-style "screen intelligence" to help Codex Readers have engaged with this next
  3. Wiki concept OpenAI Whisper Speech-to-text model family used for video transcription and realtime speech recognition. In the wiki it first appears as the transcription backend for AI video Related by gpt