Breaking

OpenAI Brings Moderation Signals Into Its Generation APIs

June 4, 2026 at 15:37 EDT

OpenAI announced on June 4, 2026, that moderation scores are now available within generation requests in both the Responses API and the Completions API. Safety signals can be obtained in the same request flow as generation, letting apps flexibly decide how to log, route, review, or block.

June 4, 2026 · OpenAI Developers

Moderation Goes Inline: Safety Signals Now Ride Along With Generation

Moderation scores are now available inside generation requests in both the Responses API and the Completions API — no more separate call. Apps get a live safety signal in the same request path, then decide their own logic: log, route, review, or block.

Before → After: API calls per generation

Before
generate + moderate

CALL 1

CALL 2

2×

After
inline signal

ONE CALL

1×

One request instead of two — moderation becomes plumbing in the same path, cutting the latency of a separate call.

FREE

Moderations endpoint stays free — no added fees for inline integration.

0–1

Per-category confidence scores — higher means greater certainty of a violation.

20MB

Image input limit on the multimodal omni-moderation-latest model.

Get a signal, then your app decides

Generate

+ inline scores

→

Log

Route

Review

Block

Categories covered: harassment · hate · self-harm · sexual · violence. Thresholds may need recalibration as the models update.

Praise

"One API call instead of two."
Safety as a live signal in the same path — not a lecture after generation.
Welcomed by teams already doing in-house risk scoring.

Skepticism

"Just another layer of middleware adding latency."
Must scoring be configured per use case?
Many users were instead awaiting 4o's return or GPT-5.6.

193 likes

31 replies

~15K views

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.