GPT-4.1 mini

+ contribute

GPT-4.1 mini is a closed-weights language model from OpenAI, released April 2025.

60.3±0.6#191 of 719 LLMsfrom 43 benchmark results across 19 sources

$ how it compares

The LLMs ranked either side of GPT-4.1 mini, out of 828 with a score, then more from OpenAI.

#	model	score	where	RAM	released
227	Nemotron-Cascade-14B-Thinking NVIDIA	61±1	local	~10GB	2025-12
228	Mellum2.1 12B-A2.5B Thinking JetBrains	60±2	local	~8.3GB	2026-09
229	GPT-4.1 mini OpenAI	60±1	api $0.70/M		2025-04
230	DeepSeek V3 0324 DeepSeek	60±1	local	~108GB	2025-03
231	Qwen3 30B-A3B Instruct 2507 Alibaba	60±1	api $0.15/M +local	~20GB	2025-07
	more from OpenAI
1	GPT-6 Astra OpenAI	87±1	api $20.00/M		2026-09
3	GPT-6.1 Sol OpenAI	85±1	api $4.00/M		2026-09
7	GPT-6 Sol OpenAI	83±1	api $4.00/M		2026-09

see all LLMs · find the ones that fit your machine

$ what it's good at

agent tasks▁▁▃▆█▆▅▅▅▄▃▁ #165 of 300 · 16% of its score · 10 tests

Beats o1 (2024-12-17), o3-mini, DeepSeek R1 0528 here, though they rank higher overall. Trails SmolLM3 3B, Jamba 1.5 Mini, Jamba 1.5 Large here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
τ²-bench Telecom lab +1	44.0% ✎	▃·▅▅▃▃··█▃▃▅ beats 4/13	very hard · avg 13%	▇▇ 4.0%	sharp · at its level · several runs
BrowserGym WebArena (AgentLab GenericAgent)	30.7% ✎	█··▄·▄█·▄··▄ beats 4/7	hard · avg 18%	▇ 2.4%	at its level
τ²-bench Retail	61.4% ✎	▁····▂▁▃▆█▃▁ beats 12/44	medium · avg 41%	▇ 2.1%	at its level
Finance Agent Benchmark v1 (537 questions, class-balanced accuracy)	20.3% ✎	█·▃▃·█▅▅▅▅▃▅ beats 11/18	very hard · avg 5.8%	▇ 2.0%	sharp
BFCL V4 (December 2025 archive)	50.4% ✎	▁▁▁▂██▅▅▆▂▂▂ beats 57/77	hard · avg 32%	▇ 1.7%	at its level
τ²-bench Airline	48.7% ✎	▁▁·▂▁▃▃▁█▅▂▃ beats 12/42	hard · avg 29%	▇ 1.4%	at its level
BFCL v3	65.2% ✎	▁▁▁▁▁▃▅█▅▄▁▁ beats 98/114	medium · avg 52%	▇ 0.7%	blunt · at its level
Galileo Agent Leaderboard v2 (action completion)	56.0% ✎	▅▃·▃▃▅▅█▃█▅▅ beats 17/19	hard · avg 27%	▇ 0.7%	at its level

2 more tests

test	result	vs the others in its run	difficulty	counts	why
Galileo Agent Leaderboard v1 (tool selection quality)	91.0% ✎	▃▃▃▃▆▅▆▅█▂▅▂ beats 25/31	easy · avg 79%	▇ 0.4%	blunt
KlinikeBench strict pass@1 (333 interactive clinical encounters, Terminus-2)	2.1% ✎	▁▁·▃▁▁▁▅▃█▅█ beats 2/27	very hard · avg 2.4%	▇ 0.4%	too hard for it

coding▁▁▁▃▆██▇▆▄▃▁ #175 of 458 * · 10% of its score · 2 tests

Beats Qwen3.5 9B, Devstral 2 (2512), Apriel 1.6 15B Thinker here, though they rank higher overall. Trails QwQ-32B-Preview, Ministral 3 8B Reasoning 2512, Phi-4-reasoning 14B here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
Aider polyglot (225 exercises, two attempts)	32.4% ✎	▂·▂▃▅▃▃█▆▄▂▂ beats 16/49	very hard · avg 8.5%	▇▇▇▇ 9.5%	sharp · at its level
HumanEval (simple-evals) lab	93.8% ✎	▆█▆▆▂··▂···▂ beats 12/15	easy · avg 84%	▇ 0.9%	typical

chat & writing▁▁▁▃▅▄▅█▆▆▇▃ #128 of 393 · 9.4% of its score · 3 tests

Beats LongCat 2.0 (1.6T-A48B), o3-mini, K-EXAONE (236B-A23B) here, though they rank higher overall. Trails LFM2.5 8B-A1B, LFM2.5 2.6B, Nemotron-Cascade-8B here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
Arena-Hard v2.0 hard prompts (style control, Gemini-2.5 judge)	46.9% ✎	▂·▃▅▅·▃█▆▂▂▂ beats 13/22	very hard · avg 12%	▇▇ 5.1%	sharp · at its level
LMArena (text)	1382.7 ✎	▁▁▁▂▃▃▃▆▆▇█▃ beats 223/379	avg 1268	▇ 3.3%	blunt
Arena-Hard v2.0 creative writing (GPT-4.1 + Gemini-2.5 judges)	28.2% ✎	▆▂▂▆▂▂▃█▂▃·▂ beats 9/22	hard · avg 19%	▇ 0.9%	sharp · at its level · noisy

math▁▂▃▅▆▅▄█▄▁▁▁ #174 of 398 · 18% of its score · 5 tests

Beats MiniMax-M2.1, Mistral Medium 3.5 (128B), GPT-4.1 here, though they rank higher overall. Trails OpenThinker3 7B, Olmo 3 7B Think, OpenThinker2 32B here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
MATH level 5	87.3% ✎	▂▂▅▆▄█▆▃▅▂▃▄ beats 74/92	hard · avg 32%	▇▇ 5.7%	sharp
FrontierMath tiers 1–3 v2	6.7% ✎	▂··▃▁▃▄▄█▅▃▃ beats 5/82	very hard · avg 1.2%	▇▇ 5.3%	sharp
OTIS mock AIME 2024–2025	44.7% ✎	▃▃▄▃▃▃▇▇█▃▂▅ beats 74/193	very hard · avg 8.5%	▇ 3.4%	sharp · at its level · noisy
MATH-500 (simple-evals) lab	81.4% ✎	▃··▅▅▃···▃·█ beats 1/9	medium · avg 61%	▇ 2.8%	sharp
MGSM (simple-evals) lab	88.2% ✎	▁····▁▅▃█▁▃▁ beats 6/16	easy · avg 82%	▇ 1.0%	blunt

vision & documents▁▁▃▂▁▃▄▂▆▇█▄ #77 of 147 · 7.0% of its score · 3 tests

Beats DeepSeek V4 Flash Vision (Exp), GPT-5.4 nano, GPT-5 mini here, though they rank higher overall.

test	result	vs the others in its run	difficulty	counts	why
CharXiv descriptive questions lab	88.4% ✎	▄▄·····▄·▄█▄ beats 4/6	easy · avg 72%	▇ 2.4%	typical
Arena vision (style controlled)	1200.8 ✎	▁▁▂▂▂▂▄▅▅▆██ beats 73/145	avg 1068	▇ 2.3%	blunt
CharXiv (reasoning questions) lab	56.8% ✎	▄··▂█··▂···▄ beats 6/9	hard · avg 34%	▇ 2.3%	at its level

web search▂▁▂▆▄█▃▃▁▃·▁ #5 of 80 * · 2.0% of its score · 1 test

Beats GPT-6 Astra, Claude Opus 5.5, Claude Fable 5.1 here, though they rank higher overall.

test	result	vs the others in its run	difficulty	counts	why
Finance Agent Benchmark v1 (537 questions, class-balanced accuracy)	20.3% ✎	█·▃▃·█▅▅▅▅▃▅ beats 11/18	very hard · avg 5.8%	▇ 2.0%	sharp

other tests19 not in a task · 39% of its score

test	result	vs the others in its run	difficulty	counts	why
GPQA Diamond (simple-evals) lab	65.0% ✎	█▃▃█·▃▃▅▃▃▃▅ beats 9/16	hard · avg 40%	▇▇▇ 7.7%	sharp · at its level
GPQA Diamond	65.8% ✎	▁▁▁▁▂▄▆▄▅▇█▅ beats 98/208	hard · avg 40%	▇▇ 6.2%	sharp · at its level
CLIcK Korean (Daekeun Kim)	72.1% ✎	▆▄▄▄█▄▂·▄█·▄ beats 10/23	medium · avg 58%	▇▇ 3.8%	typical
KMMLU-HARD (Daekeun Kim, zero-shot)	35.6% ✎	█···▁▃▁▂▃·▆▂ beats 9/23	very hard · avg 29%	▇ 3.1%	sharp
HAE-RAE 1.0 (Daekeun Kim)	78.9% ✎	▄▄▄▂▂█▂▄▂█▂▄ beats 11/22	medium · avg 63%	▇ 3.0%	typical
KCL-Essay vanilla	30.9% ✎	▁···▁▃▂█▂▃▃▂ beats 8/23	hard · avg 18%	▇ 2.3%	at its level
ARC-AGI-1 public evaluation set (ARC Prize model_baseline, pass@2)	7.2% ✎	▁·▇▃▁·█▇▁·▇▃ beats 2/27	very hard · avg 4.1%	▇ 2.1%	sharp · noisy
EQ-Bench 3 (emotional intelligence, judge Elo)	1171.0 ✎	▁·▂▁▄▅▅▃█▄▁▁ beats 26/62	avg 1044	▇ 2.1%	typical

11 more tests

test	result	vs the others in its run	difficulty	counts	why
MMLU (simple-evals, zero-shot CoT) lab	87.5% ✎	▂▂·█▄▄▂·█▂·▂ beats 9/16	easy · avg 80%	▇ 2.0%	typical
HindiGen-v1 3C3H score (Inception, Claude 3.5 Sonnet judge)	65.0% ✎	▁▁▁▁·▂·█▆█▇▁ beats 16/31	medium · avg 48%	▇ 1.5%	at its level
Open Persian LLM Leaderboard v2 average (21 tasks)	70.1% ✎	▂▄▂▃█▄▃▆▃▅▂▂ beats 74/81	medium · avg 58%	▇ 1.3%	blunt · at its level
KCL-MCQA vanilla	29.4% ✎	▂···█▂▆▆██▆▂ beats 4/23	very hard · avg 30%	▇ 1.1%	at its level
SimpleQA Verified	12.7% ✎	▁▁▃▂▃▃▆▆█▃▂▅ beats 7/78	very hard · avg 8.0%	▇ 0.8%	typical
Chess Puzzles (Epoch)	7.0% ✎	█▁▃▁▁▂▁▄▂▂▁▁ beats 51/135	very hard · avg 5.9%	▇ 0.6%	sharp · too hard for it · noisy
ARC-AGI-2 public evaluation set (ARC Prize model_baseline, pass@2)	0.0% ✎	█··▂▅▁▃▁▃·▁▃ beats 1/24	very hard · avg 0.3%	▇ 0.5%	sharp · too hard for it · noisy
Mystery Game Puzzles (Epoch)	7.0% ✎	█··▂▃▇▄█▂▃▁▁ beats 6/74	very hard · avg 9.4%	▇ 0.4%	sharp · too hard for it · noisy
SimpleQA (simple-evals) lab	16.8% ✎	▄·▁▃▁▁··█▁·▁ beats 4/14	hard · avg 18%	▇ 0.3%	at its level · noisy
DROP F1, 3-shot (simple-evals) lab	81.0% ✎	▂▂▄▆▆█·▂···▄ beats 6/16	easy · avg 78%	▇ 0.3%	blunt
Vectara hallucination leaderboard, 2025 dataset (HHEM-2.3, factual consistency rate)	97.8% ✎	▁▁▂▃▄▄█▆▃▂▂▁ beats 120/140	very easy · avg 94%	▇ 0.1%	blunt · maxed out · noisy

$ where to run it

Through an API

api · $0.70/M · $0.4 in / $1.6 out per 1M tokens · 1M context · platform.openai.com ↗ · price sourceStandard base token rate from models.dev. Provider-specific tiers, long-context surcharges, caching and promotions may differ.

$ how the score was worked out · Epoch AI — internal evaluatio… 22%, OpenAI simple-evals README be… 15%, Daekeun Kim Korean LLM benchm… 10% and 16 more

Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works

source	results	share of the score	licence
Epoch AI — internal evaluations	7	▇▇▇▇▇▇▇ 22%	CC BY 4.0
OpenAI simple-evals README benchmark results (OpenAI's own models) · lab-reported	7	▇▇▇▇▇ 15%	MIT
Daekeun Kim Korean LLM benchmark runs	3	▇▇▇ 10%	MIT
Aider polyglot benchmark runs	1	▇▇▇ 9%	Apache-2.0
OpenAI vendor-reported benchmark scores for its own models (openai.com announcement posts and Deployment Safety Hub system cards) · lab-reported	3	▇▇ 6%	Factual vendor-reported score (not licensed content)
Arena-Hard-Auto v2.0-Preview leaderboard (repository README)	2	▇▇ 6%	Apache-2.0
τ²/τ³-bench leaderboard: Sierra's own evaluation runs	3	▇▇ 6%	MIT
LMArena leaderboard dataset	2	▇▇ 6%	CC BY 4.0
Korean Canonical Legal Benchmark: Toward Knowledge-Independent Evaluation of LLMs’ Legal Reasoning Capabilities	2	▇ 3%	CC BY 4.0
ARC Prize Foundation model_baseline outputs on ARC-AGI-1/2 public evaluation sets	2	▇ 3%	MIT
ServiceNow BrowserGym leaderboard (AgentLab GenericAgent runs)	1	▇ 2%	MIT
Berkeley Function Calling Leaderboard result archive	2	▇ 2%	Apache-2.0
EQ-Bench 3 canonical leaderboard results	1	▇ 2%	MIT
Finance Agent Benchmark (Vals AI, arXiv:2508.00828v1)	1	▇ 2%	CC-BY-4.0
Inception Arabic/Hindi leaderboards (AraGen-v3, HindiGen-v1, Arabic IFEval)	1	▇ 2%	Apache-2.0
Open Persian LLM Leaderboard v2 (Part DP AI and AUT NLP Lab)	1	▇ 1%	Apache-2.0
Galileo Agent Leaderboard (v1 and v2 results)	2	▇ 1%	Apache-2.0
KlinikeBench: Evaluating Language Models Beyond Diagnostic Accuracy (arXiv:2609.38480v2)	1	▇ 0%	CC BY 4.0
Vectara hallucination leaderboard (vectara/hallucination-leaderboard repository)	1	▇ 0%	Apache-2.0
prior · pull toward the middle		▇ 0%
weighted average		= 60.3