GPT-5 mini

+ contribute

GPT-5 mini is a closed-weights language model from OpenAI, released August 2025.

67.4±0.6#109 of 719 LLMsfrom 53 benchmark results across 22 sources

$ how it compares

The LLMs ranked either side of GPT-5 mini, out of 828 with a score, then more from OpenAI.

#	model	score	where	RAM	released
122	Ornith 1.0 35B (35B-A3B) DeepReinforce	68±1	local	~23GB	2026-06
123	Hy3 preview (295B-A21B) Tencent	68±1	local	~629GB	2026-04
124	GPT-5 mini OpenAI	67±1	api $0.69/M		2025-08
125	Gemini 2.5 Pro Google	67±1	api $3.44/M		2025-06
126	Claude Opus 4 Anthropic	67±1	comparison only		2025-05
	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▁▁▃▆█▆▅▅▅▄▃▁ #121 of 300 · 37% of its score · 17 tests

Beats o3 here, though they rank higher overall. Trails SmolLM3 3B, Jamba 1.5 Large, Nova 2 Pro (Preview) here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
SWE-bench Verified +1	64.7% ✎	▁··▁·▁▃▁▄█▄▁ beats 6/29	hard · avg 19%	▇▇▇ 8.7%	sharp · at its level · several runs
τ²-bench aggregate +1	69.8% ✎	▁▁▁·▂▂▄▃▃▅█▂ beats 15/37	medium · avg 36%	▇ 3.4%	several runs
Terminal-Bench 2.0 with Terminus 2 agent (benchmark authors' runs)	24.0% ✎	▃··▃·▃▃██▅▃▅ beats 5/14	very hard · avg 5.8%	▇ 2.8%	sharp · at its level
BrowserGym WorkArena-L2 (AgentLab GenericAgent)	47.7% ✎	▃▅▃·▅▃▃··█·▃ beats 10/11	very hard · avg 3.9%	▇ 2.4%	sharp · at its level · noisy
Terminal-Bench 2.0 lab	38.2% ✎	██···██··███ beats 0/6	very hard · avg 14%	▇ 2.4%	at its level
τ²-bench Telecom lab +1	74.1% ✎	▃·▅▅▃▃··█▃▃▅ beats 6/13	hard · avg 16%	▇ 2.3%	sharp · at its level · noisy · several runs
Toolathlon (Tool Decathlon) lab	26.9% ✎	▃··▃··▃··▃█▅ beats 0/8	very hard · avg 13%	▇ 1.9%	at its level
Humanity's Last Exam with tools lab	31.6% ✎	▃·▃·▅····▅·█ beats 0/8	very hard · avg 12%	▇ 1.7%	at its level

9 more tests

test	result	vs the others in its run	difficulty	counts	why
τ²-bench Retail	73.4% ✎	█▄··▄··▄·▄·▄ beats 5/6	medium · avg 37%	▇ 1.7%	at its level
BrowserGym WorkArena-L1 (AgentLab GenericAgent)	60.6% ✎	█···█▄▄·▄█·█ beats 9/10	hard · avg 29%	▇ 1.7%	at its level
BFCL V4 (December 2025 archive)	55.5% ✎	▁▁▁▂██▅▅▆▂▂▂ beats 65/77	hard · avg 32%	▇ 1.4%	at its level
τ²-bench Airline	59.3% ✎	▄▄······▄█▄▄ beats 5/6	hard · avg 29%	▇ 1.4%	at its level
OSWorld-Verified lab	42.0% ✎	█▄·····▄·▄▄▄ beats 2/6	hard · avg 17%	▇ 1.3%	at its level
Vibe Code Bench (test split, 50 web apps)	14.2% ✎	▃··▃·▃██▃▃▃█ beats 3/14	very hard · avg 2.3%	▇ 1.2%	sharp · noisy
BrowserGym MiniWoB (AgentLab GenericAgent)	71.0% ✎	▅····▅▅▃▃·▃█ beats 10/11	medium · avg 55%	▇ 0.9%	blunt · at its level
SWE-bench Pro lab	45.7% ✎	▄·····▄·▄▄▄█ beats 0/6	hard · avg 28%	▇ 0.8%	blunt · at its level
τ²-Bench-Verified (GPT-5.1 user)	69.8% ✎	▂··▄···▃▄▄█▄ beats 3/13	medium · avg 45%	▇ 0.6%	at its level

coding▁▁▁▃▆██▇▆▄▃▁ #110 of 458 · 19% of its score · 5 tests

Beats Grok 4.3, Qwen3.5 Plus, DeepSeek V3.2 here, though they rank higher overall. Trails Llama 3.1 Nemotron Ultra 253B v1, Solar Pro 4, Laguna XS 2.1 (33B-A3B) here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
SWE-bench Verified +1	64.7% ✎	▁··▁·▁▃▁▄█▄▁ beats 6/29	hard · avg 19%	▇▇▇ 8.7%	sharp · at its level · several runs
Aider polyglot (225 exercises, two attempts)	71.6% ✎	█▄····▄··█·▄ beats 5/6	very hard · avg 7.9%	▇▇▇ 7.2%	sharp · at its level
VisEditBench strict pass rate (1,395 visualization-code editing tasks)	60.0% ✎	▄··█·▄··▄█·▄ beats 4/7	hard · avg 34%	▇ 1.5%	at its level
Vibe Code Bench (test split, 50 web apps)	14.2% ✎	▃··▃·▃██▃▃▃█ beats 3/14	very hard · avg 2.3%	▇ 1.2%	sharp · noisy
SWE-bench Pro lab	45.7% ✎	▄·····▄·▄▄▄█ beats 0/6	hard · avg 28%	▇ 0.8%	blunt · at its level

chat & writing▁▁▁▃▅▄▅█▆▆▇▃ #121 of 393 * · 3.0% of its score · 1 test

Beats LongCat 2.0 (1.6T-A48B) here, though they rank higher overall. Trails LFM2.5 2.6B, Nemotron-Cascade-8B, DeepSeek V3 0324 here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
LMArena (text)	1389.7 ✎	▁▁▁▂▃▃▃▆▆▇█▃ beats 232/379	avg 1268	▇ 3.0%	blunt

math▁▂▃▅▆▅▄█▄▁▁▁ #61 of 398 · 21% of its score · 7 tests

Beats Qwen3.6 Max Preview, Claude Sonnet 4.6, Claude Opus 4.5 here, though they rank higher overall. Trails INTELLECT-3 here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
FrontierMath tiers 1–3 v2	46.7% ✎	▂··▁▁▃▄▄█▅▃▃ beats 34/82	very hard · avg 1.2%	▇▇▇▇▇ 12%	sharp · at its level
FrontierMath tier 4 v2	12.2% ✎	▁▂▁▃█▅▂▂▁▁▁▁ beats 14/63	very hard · avg 0.1%	▇ 3.1%	sharp · too hard for it
OTIS mock AIME 2024–2025	86.7% ✎	▃▃▄▃▃▃▇▇█▃▂▅ beats 133/193	very hard · avg 8.5%	▇ 2.0%	sharp · noisy
MATH level 5	97.8% ✎	▂▂▅▆▄█▆▃▅▂▃▄ beats 91/92	hard · avg 32%	▇ 1.7%	sharp · maxed out
MathNet-Solve-Test micro-average accuracy (6,400 olympiad problems)	57.0% ✎	▃▃···█▃·▃·▃▃ beats 6/8	hard · avg 15%	▇ 1.1%	at its level
MATH-Vision	71.9% ✎	▁▁·▁··▄▁▃▇▇█ beats 10/24	medium · avg 47%	▇ 1.0%	typical
FormalProofBench (200 graduate-level Lean 4 problems, agentic)	7.0% ✎	▃▅·▃··▅▃█··▃ beats 3/10	very hard · avg 3.0%	▇ 0.5%	noisy

vision & documents▁▁▃▂▁▃▄▃▆▇█▄ #85 of 147 · 7.8% of its score · 7 tests

Beats DeepSeek V4 Flash Vision (Exp) here, though they rank higher overall. Trails GPT-4.1 mini, Qwen3.5 4B, Gemini 2.5 Flash here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
MMMU Pro lab	67.5% ✎	▅·····▃·▃·█▃ beats 1/7	hard · avg 40%	▇ 2.6%	at its level
Arena vision (style controlled)	1183.2 ✎	▁▁▂▂▂▂▄▅▅▆██ beats 61/145	avg 1068	▇ 2.0%	blunt
VisEditBench strict pass rate (1,395 visualization-code editing tasks)	60.0% ✎	▄··█·▄··▄█·▄ beats 4/7	hard · avg 34%	▇ 1.5%	at its level
MATH-Vision	71.9% ✎	▁▁·▁··▄▁▃▇▇█ beats 10/24	medium · avg 47%	▇ 1.0%	typical
Molmo2-VideoCount accuracy (val)	29.8% ✎	▃··▃·▃····█▃ beats 1/6	hard · avg 22%	▇ 0.3%	blunt · at its level
Ai2 Molmo2 human preference Elo (video/image QA, 2026)	1076.0 ✎	▄··▄··▄··▄█▄ beats 3/6	avg 975	▇ 0.2%	blunt
Molmo2-CapTest dense video captioning F1	56.6% ✎	█···█·████·█ beats 6/6	hard · avg 32%	▇ 0.1%	blunt · at its level

web search▂▁▃▆▄█▃▃▁▁·▁ #73 of 80 * · 1.7% of its score · 1 test

Trails GPT-4.1 nano, GPT-4o mini (2024-07-18), Mistral Small 3.1 24B (2503) here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
Humanity's Last Exam with tools lab	31.6% ✎	▃·▃·▅····▅·█ beats 0/8	very hard · avg 12%	▇ 1.7%	at its level

other tests18 not in a task · 25% of its score

test	result	vs the others in its run	difficulty	counts	why
GPQA Diamond	75.0% ✎	▁▁▁▁▂▄▆▄▅▇█▅ beats 112/208	hard · avg 40%	▇▇ 4.3%	sharp · at its level
KMMLU-HARD (Daekeun Kim, zero-shot)	60.6% ✎	█···▁▁▁▂▃·▆▂ beats 15/23	very hard · avg 29%	▇▇ 4.1%	sharp · at its level
Humanity's Last Exam lab	18.3% ✎	▃··▃·▃▃···▃█ beats 0/7	very hard · avg 4.4%	▇ 2.4%	typical
CLIcK Korean (Daekeun Kim)	82.2% ✎	▆▄▄▄█▄▃·▄█·▄ beats 15/23	medium · avg 58%	▇ 2.4%	typical
KCL-Essay vanilla	40.3% ✎	▁···▁▃▂█▃▃▃▂ beats 15/23	hard · avg 18%	▇ 2.3%	at its level
HAE-RAE 1.0 (Daekeun Kim)	84.4% ✎	▄▄▄▂▂█▃▄▂█▂▄ beats 12/22	medium · avg 63%	▇ 1.9%	typical
Mystery Game Puzzles (Epoch)	5.0% ✎	█··▂▃▇▄█▂▃▁▁ beats 12/74	very hard · avg 9.4%	▇ 1.2%	sharp · too hard for it · noisy
Chess Puzzles (Epoch)	30.0% ✎	█▁▁▁▁▂▁▄▃▂▁▁ beats 104/135	very hard · avg 5.9%	▇ 1.2%	sharp · noisy

10 more tests

test	result	vs the others in its run	difficulty	counts	why
KCL-MCQA vanilla	40.9% ✎	▂···█▂▆▆██▆▂ beats 11/23	very hard · avg 30%	▇ 1.2%	at its level
SimpleQA Verified	21.6% ✎	▁▁▃▂▃▃▆▆█▃▂▅ beats 14/78	very hard · avg 8.0%	▇ 1.0%	at its level
SimpleQA lab	22.0% ✎	▄▄▄▆█▆███·▂█ beats 20/32	very hard · avg 9.1%	▇ 0.8%	at its level · noisy
Multi-IF (aggregation unspecified)	85.4% ✎	▄·····▄█▄▄·▄ beats 5/6	easy · avg 66%	▇ 0.7%	blunt
SOOHAK Challenge (340 problems, Avg@3)	18.8% ✎	█·█████··█·█ beats 6/7	very hard · avg 4.1%	▇ 0.5%	typical
SOOHAK-Mini (702 problems, Avg@3)	67.1% ✎	▄··▄·▄·▄▄█·▄ beats 6/7	medium · avg 36%	▇ 0.4%	blunt · at its level
HealthBench Hard lab	40.3% ✎	▁····▁▁▂█·▃▁ beats 12/13	hard · avg 15%	▇ 0.4%	blunt · at its level · small run
Open PT-LLM Leaderboard average (9 Portuguese tasks)	83.9% ✎	▃▁▁▁▁▁▁▁▂█▅▃ beats 273/276	medium · avg 73%	▇ 0.3%	noisy
Vectara hallucination leaderboard, 2025 dataset (HHEM-2.3, factual consistency rate)	96.8% ✎	▁▁▂▃▄▄█▆▃▂▂▁ beats 99/140	very easy · avg 94%	▇ 0.1%	blunt · maxed out · noisy
Vectara hallucination leaderboard (HHEM-2.3, factual consistency rate)	87.1% ✎	▁▂▁▂▅█▂▃▄▃▁▁ beats 17/98	very easy · avg 90%	▇ 0.0%	blunt

$ where to run it

Through an API

api · $0.69/M · $0.25 in / $2 out per 1M tokens · 400k 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… 33%, OpenAI vendor-reported benchm… 14%, ServiceNow-AI/Apriel-1.6-15b-… 13% and 19 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	9	▇▇▇▇▇▇▇▇▇▇ 33%	CC BY 4.0
OpenAI vendor-reported benchmark scores for its own models (openai.com announcement posts and Deployment Safety Hub system cards) · lab-reported	8	▇▇▇▇ 14%	Factual vendor-reported score (not licensed content)
ServiceNow-AI/Apriel-1.6-15b-Thinker model card	5	▇▇▇▇ 13%	MIT
Daekeun Kim Korean LLM benchmark runs	3	▇▇▇ 8%	MIT
ServiceNow BrowserGym leaderboard (AgentLab GenericAgent runs)	3	▇▇ 5%	MIT
LMArena leaderboard dataset	2	▇ 5%	CC BY 4.0
Korean Canonical Legal Benchmark: Toward Knowledge-Independent Evaluation of LLMs’ Legal Reasoning Capabilities	2	▇ 3%	CC BY 4.0
Terminal-Bench: Benchmarking Agents on Hard, Realistic Tasks in Command Line Interfaces — benchmark authors' own agent-model trials	1	▇ 3%	CC BY 4.0
Qwen/Qwen3.5-27B model card	2	▇ 3%	apache-2.0
Qwen/Qwen3.5-122B-A10B model card	1	▇ 2%	apache-2.0
VisEditBench: Can Vision-Language Models Edit Visualization Code from Multimodal Feedback? (arXiv:2608.10408v1)	1	▇ 2%	CC BY 4.0
Berkeley Function Calling Leaderboard result archive	1	▇ 1%	Apache-2.0
Vibe Code Bench (Vals AI, arXiv:2603.04601v3)	1	▇ 1%	CC-BY-4.0
OpenAI — GPT-5 System Card (arXiv:2601.03267v2) · lab-reported	2	▇ 1%	CC BY 4.0
MathNet: a Global Multimodal Benchmark for Mathematical Reasoning and Retrieval — authors' own model runs	1	▇ 1%	CC BY 4.0
SEATauBench: Progressively Adapting Tool-Agent-User Evaluation Into Low-Resource Southeast Asian Languages (arXiv:2606.28715v2)	1	▇ 1%	CC-BY-4.0
Soohak: A Mathematician-Curated Benchmark for Evaluating Research-level Math Capabilities of LLMs — authors' own model runs	2	▇ 1%	CC BY-SA 4.0
Molmo2: Open Weights and Data for Vision-Language Models with Video Understanding and Grounding (Ai2, arXiv 2601.10611v4)	3	▇ 1%	CC BY 4.0
Amazon AGI — τ²-Bench-Verified	1	▇ 1%	MIT
FormalProofBench (Vals AI, arXiv:2603.26996v1)	1	▇ 0%	CC-BY-4.0
Open Portuguese LLM Leaderboard (archived)	1	▇ 0%	Apache-2.0
Vectara hallucination leaderboard (vectara/hallucination-leaderboard repository)	2	▇ 0%	Apache-2.0
prior · pull toward the middle		▇ 0%
weighted average		= 67.4