2026-04-23
OpenAI launches GPT-5.5 (codename Spud) with 1M context, $5/$30 pricing
GPT-5.5 (codename Spud) shipped to ChatGPT and Codex Apr 23, API Apr 24. OpenAI cites Terminal-Bench 2.0 82.7% and FrontierMath gains over Opus 4.7.
Per TechCrunch, OpenAI launched GPT-5.5 (internal codename “Spud”) and GPT-5.5 Pro. Rollout reached ChatGPT Plus / Pro / Business / Enterprise and Codex on April 23, with API access following on April 24. OpenAI cites a Terminal-Bench 2.0 score of 82.7% and FrontierMath gains versus Claude Opus 4.7 and Gemini 3.1 Pro.
API pricing per the report:
gpt-5.5— $5/M input, $30/M output, 1M-token contextgpt-5.5-pro— $30/M input, $180/M output
Practitioner note
Two things stand out in the pricing table:
- gpt-5.5 input matches Claude Opus 4.7 input ($5/M). With 1M context and stronger Terminal-Bench, this is now a head-to-head choice for agentic-coding harnesses, not a cheap-vs-expensive split.
- gpt-5.5-pro output is $180/M. Sticker shock, but for hard math / multi-step reasoning where you want one good answer instead of many cheap ones, the cost-per-correct-answer math may favor Pro. Run your own eval before assuming Pro is “too expensive.”
I’d benchmark gpt-5.5 against Opus 4.7 on your own agent harness before defaulting to either.