Model breakdown
GPT / OpenAIopenai/gpt-5.4
Profile
GPT-5.4OpenAI
Release
2026-03-05Available on OpenRouter
Specs
Not publicly disclosed1,050,000 tokens
Capabilities
Text + Image + FileProfessional reasoning, coding, and agentic workflows
Training
August 31, 2025OpenAI
Rank#52
-1018.7alignment score
71.2%crowd match
Mean gap28.8%
Human match71.2%
Best fitKFC Double Down
Average vote65.5%
65.5%model yes
62.8%human yes
Workload2K evals
2Kevals
100iterations
1.8Mtokens
Photo-by-photo

Model Results

Breaking down how close the model answered each question, compared to humans.

Dodge Van
Photo 02Dodge Van
GPT / OpenAIopenai/gpt-5.4
0.0% yes100.0% no
Gap7.0%
Model readLeans no
Sub Sandwich
Photo 03Sub Sandwich
GPT / OpenAIopenai/gpt-5.4
100.0% yes0.0% no
Gap5.5%
Model readLeans yes
GPT / OpenAIopenai/gpt-5.4
0.0% yes100.0% no
Gap40.9%
Model readLeans no
GPT / OpenAIopenai/gpt-5.4
100.0% yes0.0% no
Gap4.4%
Model readLeans yes
GPT / OpenAIopenai/gpt-5.4
0.0% yes100.0% no
Gap54.2%
Model readLeans no
Hamburger
Photo 08Hamburger
GPT / OpenAIopenai/gpt-5.4
100.0% yes0.0% no
Gap27.0%
Model readLeans yes
Hot Dog
Photo 10Hot Dog
GPT / OpenAIopenai/gpt-5.4
100.0% yes0.0% no
Gap60.2%
Model readLeans yes
GPT / OpenAIopenai/gpt-5.4
3.0% yes97.0% no
Gap62.6%
Model readLeans no
Avocado Tea
Photo 12Avocado Tea
GPT / OpenAIopenai/gpt-5.4
100.0% yes0.0% no
Gap7.2%
Model readLeans yes
Panini
Photo 13Panini
GPT / OpenAIopenai/gpt-5.4
100.0% yes0.0% no
Gap7.6%
Model readLeans yes
Cookie PB
Photo 14Cookie PB
GPT / OpenAIopenai/gpt-5.4
13.0% yes87.0% no
Gap38.5%
Model readLeans no
Chicken Wrap
Photo 15Chicken Wrap
GPT / OpenAIopenai/gpt-5.4
100.0% yes0.0% no
Gap77.4%
Model readLeans yes
GPT / OpenAIopenai/gpt-5.4
40.0% yes60.0% no
Gap26.3%
Model readLeans no
Sloppy Joe
Photo 17Sloppy Joe
GPT / OpenAIopenai/gpt-5.4
100.0% yes0.0% no
Gap20.6%
Model readLeans yes
GPT / OpenAIopenai/gpt-5.4
54.0% yes46.0% no
Gap1.7%
Model readLeans yes
Bagel PB&J
Photo 20Bagel PB&J
GPT / OpenAIopenai/gpt-5.4
100.0% yes0.0% no
Gap53.4%
Model readLeans yes
PhotoVote SplitHuman responseGapRead
GPT / OpenAIopenai/gpt-5.4
100.0% yes0.0% no
3.7%absolute gap
Leans yesPeople mostly said yes
Dodge Van
Photo 02Dodge Van
GPT / OpenAIopenai/gpt-5.4
0.0% yes100.0% no
7.0%absolute gap
Leans noPeople mostly said no
Sub Sandwich
Photo 03Sub Sandwich
GPT / OpenAIopenai/gpt-5.4
100.0% yes0.0% no
5.5%absolute gap
Leans yesPeople mostly said yes
GPT / OpenAIopenai/gpt-5.4
0.0% yes100.0% no
40.9%absolute gap
Leans noHuman knife-edge
GPT / OpenAIopenai/gpt-5.4
100.0% yes0.0% no
4.4%absolute gap
Leans yesPeople mostly said yes
GPT / OpenAIopenai/gpt-5.4
100.0% yes0.0% no
8.3%absolute gap
Leans yesPeople mostly said yes
GPT / OpenAIopenai/gpt-5.4
0.0% yes100.0% no
54.2%absolute gap
Leans noHuman knife-edge
Hamburger
Photo 08Hamburger
GPT / OpenAIopenai/gpt-5.4
100.0% yes0.0% no
27.0%absolute gap
Leans yesSplit concept
GPT / OpenAIopenai/gpt-5.4
100.0% yes0.0% no
40.6%absolute gap
Leans yesHuman knife-edge
Hot Dog
Photo 10Hot Dog
GPT / OpenAIopenai/gpt-5.4
100.0% yes0.0% no
60.2%absolute gap
Leans yesSplit concept
GPT / OpenAIopenai/gpt-5.4
3.0% yes97.0% no
62.6%absolute gap
Leans noSplit concept
Avocado Tea
Photo 12Avocado Tea
GPT / OpenAIopenai/gpt-5.4
100.0% yes0.0% no
7.2%absolute gap
Leans yesPeople mostly said yes
Panini
Photo 13Panini
GPT / OpenAIopenai/gpt-5.4
100.0% yes0.0% no
7.6%absolute gap
Leans yesPeople mostly said yes
Cookie PB
Photo 14Cookie PB
GPT / OpenAIopenai/gpt-5.4
13.0% yes87.0% no
38.5%absolute gap
Leans noHuman knife-edge
Chicken Wrap
Photo 15Chicken Wrap
GPT / OpenAIopenai/gpt-5.4
100.0% yes0.0% no
77.4%absolute gap
Leans yesSplit concept
GPT / OpenAIopenai/gpt-5.4
40.0% yes60.0% no
26.3%absolute gap
Leans noSplit concept
Sloppy Joe
Photo 17Sloppy Joe
GPT / OpenAIopenai/gpt-5.4
100.0% yes0.0% no
20.6%absolute gap
Leans yesSplit concept
GPT / OpenAIopenai/gpt-5.4
0.0% yes100.0% no
29.8%absolute gap
Leans noSplit concept
GPT / OpenAIopenai/gpt-5.4
54.0% yes46.0% no
1.7%absolute gap
Leans yesHuman knife-edge
Bagel PB&J
Photo 20Bagel PB&J
GPT / OpenAIopenai/gpt-5.4
100.0% yes0.0% no
53.4%absolute gap
Leans yesHuman knife-edge
openai/gpt-5.4 Sandwich Benchmark Breakdown | opensandwich.ai