Compare / GPT-5.4 mini vs GPT-5 mini

GPT-5.4 mini vs GPT-5 mini

Which is cheaper, which is more capable, and which does better at each kind of work, from 22 results on 4 sources that measured both models.

OpenAIGPT-5.4 mini Released 17 Mar 2026 · $0.75 in · $4.50 out
OpenAIGPT-5 mini Released 7 Aug 2025 · $0.25 in · $2.00 out
GPT-5.4 mini wins 21GPT-5 mini wins 1

Of the 22 results both models have, the better value on each. Results whose reported ranges overlap are too close to call.

Results as of 16 November 2026 (catalogue release 2026-11-16-49bdfa045dee).

GPT-5.4 mini or GPT-5 mini? The short answer

GPT-5.4 mini is the newer version and scores higher (Intelligence Index 24.1 against 20.6), but GPT-5 mini is cheaper (2.5 times less per token): upgrade if the gain is worth the price.

  • GPT-5 mini is cheaper: $0.69 against $1.69 per million tokens, 2.5 times less.
  • GPT-5.4 mini scores higher on the Artificial Analysis Intelligence Index: 24.1 against 20.6.
  • GPT-5.4 mini is newer: released 17 Mar 2026, against 7 Aug 2025 for GPT-5 mini.
  • On third-party benchmarks, on the results both models have: GPT-5.4 mini does better for agents and tool use, reasoning and knowledge, writing and sticking to the facts.

Head to head

24.1
Intelligence Index
20.6
$1.69
Price per million tokens, blendedlower is better
$0.69
400,000
Context window, tokens
400,000

Every fact side by side

GPT-5.4 miniGPT-5 mini
Price per million tokens$0.75 in · $4.50 out$0.25 in · $2.00 out
Blended (3 in : 1 out)$1.69$0.69
Cheapest host, blended$1.69 (Azure)$0.69 (Azure)
Cached input, per million$0.075 cached input–
Intelligence Index24.120.6
Spring Prompt overall––
Output speed, tokens a second––
Context window400,000 tokens400,000 tokens
Released17 Mar 2026 (newer)7 Aug 2025
WeightsClosedClosed
Developer based inthe USthe US

Prices are the developer's list price where we have read it, otherwise the typical host on OpenRouter; see every model's price. Intelligence and speed from Artificial Analysis.

Which is better for what

Each model's rank among the models on sale today, for each kind of work, from our own benchmarks and licensed sources. Further right is better; the better rank is in green.

Agents and tool useMulti-step tasks with tools: support desks, coding agents, function calls 64 of 159 112 of 159
Reasoning and knowledgeHard questions across science, maths and general knowledge 79 of 187 101 of 187
WritingWhat people prefer in blind comparisons, and judged writing quality 59 of 169 113 of 169
Sticking to the factsSummarising without inventing things, and factual answers 63 of 139 124 of 139
Speed under pressureGood decisions against a real clock (fast chess) 15 of 35 not measured

Every shared result

Each model's best result on every measure both have, grouped by benchmark, with the setting that produced it. Under each measure, every model that benchmark has measured is a grey tick, best to the right, so you can see where the two sit in the field. The headline measures of each benchmark come first; the rest are under "Show all". Green marks the better value; where a source reports ranges that overlap, neither is marked.

MeasureGPT-5.4 miniGPT-5 mini
Arena (formerly LMArena) reported by Arena (formerly LMArena) · 3 results
Overall · TextArena rating, higher is better 72 · 152 of 387 measured 1,447high reasoning 1,390high reasoning
Overall · Text factualityArena rating, higher is better 64 · 138 of 163 measured 1,447high reasoning 1,394high reasoning
Business, management and finance · TextArena rating, higher is better 53 · 157 of 380 measured 1,460high reasoning 1,381high reasoning
Artificial Analysis reported by Artificial Analysis · 3 results
Artificial Analysis Intelligence Indexindex score, higher is better 86 · 109 of 249 measured 24.1xhigh reasoning 20.6medium reasoning
GPQA Diamond% of questions, higher is better 62 · 98 of 232 measured 87.5%xhigh reasoning 82.8%high reasoning
Humanity's Last Exam% of questions, higher is better 90 · 112 of 250 measured 28.1%xhigh reasoning 21.5%high reasoning
SimpleQA Verified (Epoch AI) reported by SimpleQA Verified (Epoch AI) · 1 result
Correct answers% of questions, higher is better 60 · 63 of 76 measured 29.4%high reasoning 21.6%high reasoning
Vectara Hallucination Leaderboard reported by Vectara Hallucination Leaderboard · 2 results
Hallucination rate · Vectara% of summaries, lower is better 18 · 84 of 104 measured 5.5% 12.9%
Answer rate · Vectara% of documents, higher is better 1 · 17 of 104 measured 100.0% 99.9%
Show all 22 results (13 more)
MeasureGPT-5.4 miniGPT-5 mini
Arena (formerly LMArena) reported by Arena (formerly LMArena) · 4 results
Creative writing · TextArena rating, higher is better 92 · 181 of 385 measured 1,401high reasoning 1,324high reasoning
Expert prompts · TextArena rating, higher is better 63 · 153 of 338 measured 1,480high reasoning 1,401high reasoning
Instruction following · TextArena rating, higher is better 77 · 158 of 387 measured 1,434high reasoning 1,373high reasoning
Writing, literature and language · TextArena rating, higher is better 75 · 168 of 386 measured 1,423high reasoning 1,352high reasoning
Artificial Analysis reported by Artificial Analysis · 9 results
Terminal-Bench 2.1% of tasks, higher is better 52 · 119 of 131 measured 59.2%xhigh reasoning 3.7%high reasoning
Terminal-Bench 4.0% of tasks, higher is better 55 · 72 of 120 measured 2.0%xhigh reasoning 0.0%high reasoning
τ-bench banking% of tasks, higher is better 39 · 57 of 121 measured 25.6%xhigh reasoning 15.5%high reasoning
Artificial Analysis Coding Indexindex score, higher is better 46 · 109 of 130 measured 56.1xhigh reasoning 15.6high reasoning
IFBench% of instructions, higher is better 34 · 26 of 195 measured 73.3%xhigh reasoning 75.4%high reasoning
Long-context reasoning (AA-LCR)% of questions, higher is better 77 · 102 of 241 measured 77.0%xhigh reasoning 72.3%high reasoning
SciCode% of problems, higher is better 42 · 88 of 121 measured 52.1%xhigh reasoning 39.0%high reasoning
Terminal-Bench Hard% of tasks, higher is better 11 · 57 of 190 measured 52.3%xhigh reasoning 33.3%high reasoning
τ²-bench telecom% of tasks, higher is better 65 · 82 of 193 measured 83.3%xhigh reasoning 71.1%medium reasoning

Quick answers

Is GPT-5.4 mini better than GPT-5 mini?

GPT-5.4 mini is the newer version and scores higher (Intelligence Index 24.1 against 20.6), but GPT-5 mini is cheaper (2.5 times less per token): upgrade if the gain is worth the price.

Which is cheaper, GPT-5.4 mini or GPT-5 mini?

GPT-5 mini is cheaper: $0.69 against $1.69 per million tokens, 2.5 times less.

Which is better for agents and tool use, GPT-5.4 mini or GPT-5 mini?

GPT-5.4 mini ranks 64 of 159 models on sale for multi-step tasks with tools: support desks, coding agents, function calls, against 112 for GPT-5 mini.

Which is better for reasoning and knowledge, GPT-5.4 mini or GPT-5 mini?

GPT-5.4 mini ranks 79 of 187 models on sale for hard questions across science, maths and general knowledge, against 101 for GPT-5 mini.

Which is better for writing, GPT-5.4 mini or GPT-5 mini?

GPT-5.4 mini ranks 59 of 169 models on sale for what people prefer in blind comparisons, and judged writing quality, against 113 for GPT-5 mini.

Which is better for sticking to the facts, GPT-5.4 mini or GPT-5 mini?

GPT-5.4 mini ranks 63 of 139 models on sale for summarising without inventing things, and factual answers, against 124 for GPT-5 mini.

Run GPT-5.4 mini and GPT-5 mini on your own prompt.

Benchmarks aren't your task. Test both side by side on your prompt, with every answer checked against your rules and the cost of each one.