Fact check
AI analysis“As of Saturday evening, September 25th, all training, evaluation, and inference with tool-use remain paused.”
Reasoning
Primary sources (OpenAI blog, Sam Altman tweet, The Verge, Reuters) state that all training, evaluation, and inference involving tool-use were paused as of September 25 evening. However, an official OpenAI clarification tweet later asserts that only evaluation is paused, with training and inference continuing. The contradictory primary statements prevent a definitive verification of the claim.
On confidence: Evidence includes both supporting statements and a direct correction from OpenAI, creating uncertainty about the current status.
Important context
The correction tweet from OpenAI modifies the original announcement, indicating that the pause may not apply to training and inference. Timing of the correction relative to the claim is crucial for determining the current state.
Evidence
Supporting (4)
- Tier 1 — Primary sourceOpenAI pauses training, evaluation, and inference with tool-use
Effective Saturday evening, September 25th, we have paused all training, evaluation, and inference that involve tool-use for our most capable models.
- Tier 1 — Primary sourceSam Altman on Twitter
All tool-use training, evaluation, and inference are paused as of tonight (Sept 25).
- Tier 2 — Independent reportingindependent originOpenAI pauses its most capable models' tool-use
OpenAI announced on Saturday that it has halted all training, evaluation, and inference involving tool-use for its latest models.
- Tier 2 — Independent reportingindependent originOpenAI pauses tool-use training, evaluation, and inference
The company said it would pause all tool-use activities as of Saturday evening, September 25.
Contradicting (1)
- Tier 1 — Primary sourceOpenAI clarification tweet
Correction: we have not paused tool-use training; only evaluation is paused. Training and inference continue as normal.
Limitations
Without exact timestamps and confirmation of which statement supersedes the other, we cannot conclusively verify the claim. Further up‑to‑date confirmation from OpenAI would be needed.
- Last verified:
- Sep 26, 2026, 5:08 PM CDT
- Pipeline:
- 0.1.0
- Claim type:
- Factual
Where this claim appeared
OpenAI pauses training of its ‘most capable models’The Verge