OpenAI SDK 7.32.0 and 3.28.0: latest versions, API diff and gotchas

OpenAI SDK 7.32.0 (2026-10-09 20:41 UTC)

• Add agent environment suspension and expiration events and webhooks (agent.session.environment.suspended, agent.session.environment.expired). (source)
• Preserve detailed token usage (prompt_tokens_details, completion_tokens_details) in runner totalUsage(). (source)

OpenAI SDK 7.31.0 (2026-10-09 04:03 UTC)

• Add prewarmed hosted environments via client.beta.agents.environments and separate existing references from inline configurations. (source)

OpenAI SDK 7.30.1 (2026-10-08 08:51 UTC)

• Correct custom voice creation parameters to require consent recordings and audio samples, dropping prompt-based creation. (source)
• Parse structured text only in final message phases, leaving non-final commentary unparsed. (source)
• Safely decode valid UTF-8 percent escapes in URL filename inference. (source)

OpenAI SDK 7.30.0 (2026-10-06 21:50 UTC)

• Add standalone Decisions support via client.decisions.create for evaluating classification and scoring questions against shared input. (source)

OpenAI SDK 7.29.0 (2026-10-06 18:50 UTC)

• Support tool search in typed function helpers via defer_loading: true in zodResponsesFunction. (source)
• Observe local tool failures and propagate error details during agent turn execution. (source)
• Run local tools during session creation streams. (source)

OpenAI SDK 7.30.0 → 7.32.0

0 exports added, 0 removed, 0 exports with public members added or removed; 12 declarations changed in any way.

Method: bin/tool-snapshot installed both versions in a throwaway Linux sandbox and listed every export of every package entry point from the type declarations with the TypeScript compiler API (921 and 921 exported symbols), then bin/tool-diff compared them. In the tables, "command" is an exported symbol and "flags" are its public members (function parameters in parentheses).

• Sandbox diff from 7.30.0 to 7.32.0. Total exported symbols remained 921 across all entry points, while 12 type declarations were updated.
• 7.31.0 added prewarmed hosted environment management (client.beta.agents.environments.create, .retrieve, .list) and tightened EnvironmentParam so passing environment_id prohibits inline configuration.
• 7.32.0 added lifecycle events and webhooks for suspended and expired environments (AgentSessionEnvironmentSuspendedEvent, AgentSessionEnvironmentExpiredEvent).
• 7.30.1 dropped VoiceCreateParams.Prompt from voice creation types, enforcing consent and audio sample uploads.
• 7.32.0 runner totalUsage() preserves nested prompt_tokens_details and completion_tokens_details instead of collapsing to top-level totals.

OpenAI SDK 7.25.0 → 7.30.0

8 exports added, 0 removed, 5 namespaces with public members or options added; 154 repository files changed across the October 2026 release cycle.

Method: Built from source (not run): compared GitHub tags v7.25.0 to v7.30.0 across 32 commits and 154 touched files, inspecting TypeScript declarations in src/client.ts, src/resources/decisions.ts, src/resources/audio/voices.ts, src/resources/beta/agents/sessions/traces.ts, src/helpers/zod.ts, and src/resources/shared.ts. In the tables, "command" is an exported symbol or API namespace and "flags" are public properties, methods, and options.

Commands added:

• openai Decisions
• openai DecisionCreateParams
• openai DecisionInputImage
• openai DecisionInputMessage
• openai DecisionInputPart
• openai DecisionInputText
• openai Audio.Voices
• openai Audio.VoiceCreateParams

• Source diff from v7.25.0 (2026-09-29) to v7.30.0 (2026-10-06). 7.30.0 added the Decisions client namespace (client.decisions) and top-level export OpenAI.Decisions.
• 7.28.0 added client.audio.voices for creating custom voice samples from audio or synthetic prompts.
• 7.26.0 and 7.27.0 added beta agent trace exploration (traces.list) returning OTLP spans and artifact handling (artifacts.download).
• 7.29.0 added defer_loading?: boolean to zodResponsesFunction options for tool search in responses.
• Chat Completions added support for detail: 'original' alongside 'low', 'high', and 'auto' on image URL parts.

Prewarmed hosted environments prohibit inline configuration (OpenAI SDK 7.31.0)

import OpenAI from "openai";
import type { EnvironmentParam } from "openai/resources/beta/agents/agents";

// In 7.32.0, mixing prewarmed environment_id with inline configuration is prohibited
const env: EnvironmentParam = {
  type: "openai_hosted",
  environment_id: "env_test_123",
  files: [{ file_id: "file_abc", path: "test.txt", type: "file_id" }],
};

test-env-mixed.ts(8,3): error TS2322: Type '{ file_id: string; path: string; type: string; }[]' is not assignable to type 'undefined'.

Fix: Supply either an existing environment reference ({ type: 'openai_hosted', environment_id }) or an inline configuration ({ type: 'openai_hosted', files: [...] }), not both. EnvironmentParamPrewarmed sets files?: never and environment_template_id?: never. (source)

Custom voice creation requires consent audio samples; prompt voices removed from SDK types (OpenAI SDK 7.30.1)

import OpenAI from "openai";
const client = new OpenAI({ apiKey: "test" });
async function main() {
  await client.audio.voices.create({
    type: "prompt",
    name: "narrator",
    prompt: "calm voice",
  });
}

test-voice.ts(5,5): error TS2322: Type '"prompt"' is not assignable to type '"audio_sample"'.

Fix: Create custom voices using audio_sample and consent recording IDs. Text prompt voice creation (type: 'prompt') was removed from VoiceCreateParams in 7.30.1. (source)

Passing transport options in the query argument of `traces.list` throws TypeError (OpenAI SDK 7.26.0)

import OpenAI from "openai";
const client = new OpenAI({ apiKey: "mock-key" });
await client.beta.agents.sessions.traces.list("session_123", {
  limit: 10,
  timeout: 5000,
});

TypeError: Query parameters and request options must be passed as separate arguments.

Fix: Pass query parameters (after, limit, order) as the second argument, and client request options (headers, timeout, signal) as the third argument: client.beta.agents.sessions.traces.list(sessionID, { limit: 10 }, { timeout: 5000 }). (source)

Missing credentials throws during client initialization (OpenAI SDK 7.0.0)

import OpenAI from "openai";
delete process.env.OPENAI_API_KEY;
const client = new OpenAI();

Error: Missing credentials. Please pass an `apiKey`, `workloadIdentity`, `adminAPIKey`, or set the `OPENAI_API_KEY` or `OPENAI_ADMIN_KEY` environment variable.

Fix: Pass apiKey explicitly to the constructor (new OpenAI({ apiKey })) or set the OPENAI_API_KEY environment variable before initializing the client. (source)

Beta agents environments API returns 401 without actor biscuit (OpenAI SDK 7.31.0)

import OpenAI from "openai";
const client = new OpenAI({ apiKey: "mock-api-key" });
await client.beta.agents.environments.retrieve("env_123");

Error: 401 A valid actor biscuit is required.

Fix: Beta Agents environment management requires an authorized OpenAI account and environment authentication biscuit (OpenAI-Beta: agents=v1). Ensure credentials and feature access are provisioned. (source)

2026-10: Node 7.26.0 to 7.32.0; Python 3.23.0 to 3.28.0 (8 versions to 2026-10-10)

• 7.32.0 and 3.28.0 (2026-10-09): agent environment suspension and expiration events, detailed token usage in runner totalUsage(). (source)
• 7.31.0 and 3.27.0 (2026-10-09): prewarmed hosted environments (client.beta.agents.environments). (source)
• 7.30.1 and 3.26.1 (2026-10-08): custom voice parameters restricted to consent audio, final-phase responses parsing, safe URL filename decoding. (source)
• 7.30.0 and 3.26.0 (2026-10-06): standalone Decisions API (client.decisions.create). (source)
• 7.29.0 (2026-10-06): tool search support in typed functions (defer_loading: true). (source)
• 7.28.0 (2026-10-04): custom voice creation (client.audio.voices.create). (source)
• 7.26.0 and 7.27.0 (2026-10-01): agent traces, final stream result collection, artifact downloads. (source)

2026-09: Node 7.9.0 to 7.25.0; Python 3.7.0 to 3.22.1 (18 versions)

• 7.23.0 (2026-09-23): GCP external storage support, GPT-Rosalind research model. (source)
• 7.22.0 (2026-09-22): GPT-6 Sol and Luna model identifiers. (source)
• 7.21.0 (2026-09-22): session environment reset events, response management resources. (source)

2026-08: Node 7.4.0 to 7.8.0; Python 3.0.0 to 3.6.0 (5 versions)

• Python 3.0.0 (2026-08-12): major Python release migrating to HTTPX2 by default. (source)
• Node 7.7.0 (2026-08-20): updated Realtime session audio format options and tool choice controls. (source)

2026-07: Node 6.46.0 to 7.3.0 (major 7.0.0 on 2026-07-27); Python 2.48.0 to 2.53.0 (8 versions)

• Node 7.0.0 (2026-07-27): major Node SDK release dropping Node 18/20 and requiring Node.js 22 LTS or later. (source)
• Node 6.49.0 (2026-07-25): final stable release on the v6.x maintenance line. (source)

2026-06: Node 6.40.0 to 6.45.0; Python 2.42.0 to 2.47.0 (6 versions)

• Node 6.45.0 (2026-06-25): added Structured Outputs support with Standard Schema integration. (source)
• Node 6.42.0 (2026-06-12): enhanced streaming events and WebSocket automatic reconnect backoff. (source)

2026-05: Node 6.36.0 to 6.39.1; Python 2.37.0 to 2.41.0 (5 versions)

• Node 6.39.0 (2026-05-28): fine-tuning checkpoint evaluation helpers and batch validation metrics. (source)
• Node 6.37.0 (2026-05-15): vector store expiration controls and file batch status callbacks. (source)