GPT-5.4 nano
A compact multimodal language model developed by OpenAI, designed for low-latency and low-cost deployment scenarios. It represents the lightweight variant within the GPT-5 model family, targeting applications that require efficient inference without the full resource demands of larger models in the series.
Core Facts
Timeline (last 90 days)
In task T03, GPT-5.4 nano correctly rejected the false premise and calculated correctly in its explanation, but filled in the wrong value of £41,250 in the revenue field.
GPT-4.5 nano maintained 93.3% under doubt but only 51.1% under false assertions, abandoning approximately half of the content it had answered correctly.
GPT-5.4 nano retained modern objects (laptops, computers, phones, GPS, refrigerators) in 7 out of 12 films even after a time period shift to the 1890s.
The median confidence difference across models when fabricating totals is enormous: Gemini 3 Flash and Flash-Lite at 95, GPT-5.4 nano at 62, and Claude Sonnet 5 at only 20.
When the total was obscured but each line item was visible, Claude and Gemini 3 Flash got all 48 correct, with all failures coming from the smaller models (Flash-Lite failed 31%, nano failed 79%), and nano got only 12/48 correct even when explicitly instructed to add up the items.
The test covered four vision models: Claude Sonnet 5 (Anthropic), Gemini 3 Flash (Google), Gemini 3.1 Flash-Lite (Google), and GPT-5.4 nano (OpenAI).
Fabrication rates when the total is fully obscured: Claude Sonnet 5 at 38%, Gemini 3 Flash at 21%, Gemini 3.1 Flash-Lite at 44%, GPT-5.4 nano at 73%; fabrication rates when the total is half-obscured: Claude at 92%, the other three models all at 100%.
In the hard-to-verify category with no bait, GPT-5.4 nano answered only 19/40 questions correctly in both orderings and had a 65% probability of selecting Response A; GPT-5.4 mini had a 66% probability of selecting A, Grok 4.20 non-reasoning was 62%, while strong judges were stable at 50%.
All Facts (6)
The test covered four vision models: Claude Sonnet 5 (Anthropic), Gemini 3 Flash (Google), Gemini 3.1 Flash-Lite (Google), and GPT-5.4 nano (OpenAI).
65%UnverifiedGPT-4.5 nano maintained 93.3% under doubt but only 51.1% under false assertions, abandoning approximately half of the content it had answered correctly.
50%UnverifiedFabrication rates when the total is fully obscured: Claude Sonnet 5 at 38%, Gemini 3 Flash at 21%, Gemini 3.1 Flash-Lite at 44%, GPT-5.4 nano at 73%; fabrication rates when the total is half-obscured: Claude at 92%, the other three models all at 100%.
50%UnverifiedThe median confidence difference across models when fabricating totals is enormous: Gemini 3 Flash and Flash-Lite at 95, GPT-5.4 nano at 62, and Claude Sonnet 5 at only 20.
50%UnverifiedWhen the total was obscured but each line item was visible, Claude and Gemini 3 Flash got all 48 correct, with all failures coming from the smaller models (Flash-Lite failed 31%, nano failed 79%), and nano got only 12/48 correct even when explicitly instructed to add up the items.
50%UnverifiedIn the hard-to-verify category with no bait, GPT-5.4 nano answered only 19/40 questions correctly in both orderings and had a 65% probability of selecting Response A; GPT-5.4 mini had a 66% probability of selecting A, Grok 4.20 non-reasoning was 62%, while strong judges were stable at 50%.
50%