OPENAI PUB_DATE: 2026.05.07

OPENAI SHIFTED DEFAULTS: GPT-5.5 INSTANT ROLLS OUT, AGENTS JS NOW DEFAULTS TO GPT-5.4-MINI, AWS BEDROCK PATH OPENS

OpenAI changed defaults across ChatGPT and the Agents SDK this week, which can silently shift behavior and costs if you don’t pin models. ChatGPT now uses [GPT...

OpenAI shifted defaults: GPT-5.5 Instant rolls out, Agents JS now defaults to gpt-5.4-mini, AWS Bedrock path opens

OpenAI changed defaults across ChatGPT and the Agents SDK this week, which can silently shift behavior and costs if you don’t pin models.

ChatGPT now uses GPT‑5.5 Instant with higher factuality and more concise answers. API callers using the chat alias may feel this too via the rolling chat‑latest target, which community threads are already probing for access and routing details (pricing/alias Q&A; routing inconsistency report).

Separately, the JS Agents SDK v0.10.0 changed its default model to gpt‑5.4‑mini (with reasoning.effort="none" and verbosity="low") and added controls like maxTurns=null and local tool concurrency. Python SDK updates this week are minor fixes (2.35.0, 2.35.1). For enterprise integration, OpenAI models and Managed Agents arrived on Amazon Bedrock (limited preview), providing a governed path to run OpenAI inside AWS.

[ WHY_IT_MATTERS ]
01.

Implicit defaults changed; unpinned agents or aliases can drift in quality, cost, and latency without any code change.

02.

AWS Bedrock access unlocks enterprise guardrails and procurement paths for teams standardized on AWS.

[ WHAT_TO_TEST ]
  • terminal

    Pin explicit models and compare outputs, token use, and latency vs the new defaults (gpt-5.4-mini and chat-latest).

  • terminal

    Stress-test Agents runs with maxTurns=null and tune toolExecution.maxFunctionToolConcurrency for throughput vs rate limits.

[ BROWNFIELD_PERSPECTIVE ]

Legacy codebase integration strategies...

  • 01.

    Audit all services that rely on model aliases or SDK defaults; set OPENAI_DEFAULT_MODEL or explicit per-agent models.

  • 02.

    Add canaries to detect routing or output drift and gate rollouts if chat-latest changes under the hood.

[ GREENFIELD_PERSPECTIVE ]

Fresh architecture paradigms...

  • 01.

    Design configs to always pin model, reasoning effort, and verbosity; treat aliases as opt-in experiments only.

  • 02.

    If you’re on AWS, explore Bedrock integration early to align with org-wide identity, logging, and data residency.

Enjoying_this_story?

Get daily OPENAI + SDLC updates.

  • Practical tactics you can ship tomorrow
  • Tooling, workflows, and architecture notes
  • One short email each weekday

FREE_FOREVER. TERMINATE_ANYTIME. View an example issue.

GET_DAILY_EMAIL
AI + SDLC // 5 MIN DAILY