GPT-5.4 vs Step 5 Preview
As of Oct 15, 2026, GPT-5.4 leads Step 5 Preview in 4 of 5 shared categories: General LLM, Reasoning, Long context, and Factuality.
GPT-5.4
It used to be. Was SOTA in General LLM.
Step 5 Preview
#33 Best current rank: #33 in Coding › Web dev, as of Oct 15, 2026.
| Category | GPT-5.4 | Step 5 Preview |
|---|---|---|
| General LLM | #30 · Was SOTA · General LLM Margin 15.957446808510639 LMArena leaderboard dataset · Oct 13, 2026 | #77 · #77 in General LLM Margin 40.95744680851064 LMArena leaderboard dataset · Oct 13, 2026 |
| Reasoning | #32 · #32 in Reasoning Margin 17.55319148936171 LMArena leaderboard dataset · Oct 13, 2026 | #63 · #63 in Reasoning Margin 34.04255319148936 LMArena leaderboard dataset · Oct 13, 2026 |
| Math | #9 · #9 in Math Margin 17.777777777777757 Epoch AI benchmarking hub · Mar 5, 2026 | #— · Not enough data Margin — LMArena leaderboard dataset · Oct 13, 2026 |
| Science | #16 · #16 in Science Margin — Epoch AI benchmarking hub · Mar 5, 2026 | — |
| Coding | #— · Not enough data Margin — LMArena leaderboard dataset · Oct 13, 2026 | #— · Not enough data Margin — LMArena leaderboard dataset · Oct 13, 2026 |
| Coding › Web dev | #60 · #60 in Coding › Web dev Margin 47.2 LMArena leaderboard dataset · Oct 13, 2026 | #33 · #33 in Coding › Web dev Margin 25.599999999999994 LMArena leaderboard dataset · Oct 13, 2026 |
| Long context | #40 · #40 in Long context Margin 21.276595744680847 LMArena leaderboard dataset · Oct 13, 2026 | #68 · #68 in Long context Margin 36.170212765957444 LMArena leaderboard dataset · Oct 13, 2026 |
| Vision | #— · Not enough data Margin — LMArena leaderboard dataset · Oct 9, 2026 | #— · Not enough data Margin — LMArena leaderboard dataset · Oct 9, 2026 |
| Factuality | #6 · Was SOTA · Factuality Margin 4.0697674418604635 LMArena leaderboard dataset · Oct 13, 2026 | #69 · #69 in Factuality Margin 40.69767441860465 LMArena leaderboard dataset · Oct 13, 2026 |
| Abstraction (ARC) | #10 · #10 in Abstraction (ARC) Margin — Epoch AI benchmarking hub · Mar 5, 2026 | — |
GPT-5.4: $5.625 per 1M blended tokens.
Step 5 Preview: $1.4249999999999998 per 1M blended tokens.
GPT-5.4: closed weights.
Step 5 Preview: closed weights.
General LLM: As of Oct 15, 2026, GPT-5.4 leads Step 5 Preview in General LLM at #30 against #77.
Reasoning: As of Oct 15, 2026, GPT-5.4 leads Step 5 Preview in Reasoning at #32 against #63.
Coding › Web dev: As of Oct 15, 2026, Step 5 Preview leads GPT-5.4 in Coding › Web dev at #33 against #60.
Long context: As of Oct 15, 2026, GPT-5.4 leads Step 5 Preview in Long context at #40 against #68.
Factuality: As of Oct 15, 2026, GPT-5.4 leads Step 5 Preview in Factuality at #6 against #69.
GPT-5.4 or Step 5 Preview?
As of Oct 15, 2026, GPT-5.4 leads Step 5 Preview in 4 of 5 shared categories: General LLM, Reasoning, Long context, and Factuality.
GPT-5.4 or Step 5 Preview for General LLM?
As of Oct 15, 2026, GPT-5.4 leads Step 5 Preview in General LLM at #30 against #77.
GPT-5.4 or Step 5 Preview for Reasoning?
As of Oct 15, 2026, GPT-5.4 leads Step 5 Preview in Reasoning at #32 against #63.
GPT-5.4 or Step 5 Preview for Coding › Web dev?
As of Oct 15, 2026, Step 5 Preview leads GPT-5.4 in Coding › Web dev at #33 against #60.