Frontier quality at a fraction of the cost

Four benchmarks, one of them independent. Each result names the model it is measured against.

Read the methodology

Enterprise testClaude Sonnet 5’s quality at 3.5% of its costBalanced, 0.952 against 0.949, on 426 held-out business prompts
RouterArenaMore accurate than GPT-5, at 6% of its cost75.5% against 74.0% correct, on an independent benchmark
HLE13.6 points above gpt-5.5, at 45% of its costQuality-first, 54.4% against 40.8% on Humanity’s Last Exam
GPQA DiamondWithin 3.7 points of gpt-5.5, at 2.4% of its costBalanced, 88.9% against 92.6% on graduate-level science

Results

Claude Sonnet 5’s quality at 3.5% of its cost

426 held-out business prompts · 15 industries · English and German · pre-registered · prices of 2 October 2026

92.7%
of gpt-5.5’s quality at 1.9% of its cost
Cost-first
97.5%
of gpt-5.5’s quality at 2.5% of its cost
Balanced (default)
99.1%
of gpt-5.5’s quality at 28% of its cost
Quality-first

Artemis presetOne model answering every requestBest score at each cost

Artemis · Cost-first0.905
$104 per 100,000 requests · quality (0 to 1)
DeepSeek-V4-Flash0.943
$130 per 100,000 requests · quality (0 to 1)
Artemis · Balanced0.952
$141 per 100,000 requests · quality (0 to 1)
Artemis · Quality-first0.967
$1,558 per 100,000 requests · quality (0 to 1)
Claude Sonnet 50.949
$3,988 per 100,000 requests · quality (0 to 1)
gpt-5.50.977
$5,596 per 100,000 requests · quality (0 to 1)

Show as a table

Quality and cost per 100,000 requests for every setup on the held-out test, cheapest first
Setup	Quality (0 to 1)	Per 100k	Cost against gpt-5.5
ministral-8b	0.687	$87	−98.4%
Nemotron-3-Nano	0.775	$91	−98.4%
Artemis, Cost-first	0.905	$104	−98.1%
DeepSeek-V4-Flash	0.943	$130	−97.7%
Artemis, Balanced (default)	0.952	$141	−97.5%
Qwen3-235B	0.853	$142	−97.5%
gpt-5.6-luna	0.962	$189	−96.6%
MiniMax-M3	0.938	$507	−90.9%
gemini-3.8-flash	0.909	$1,307	−76.7%
Artemis, Quality-first	0.967	$1,558	−72.1%
Kimi-K2.6	0.951	$2,398	−57.2%
Claude Sonnet 5	0.949	$3,988	−28.7%
gpt-5.5	0.977	$5,596	—

Every preset met its quality target

Artemis presets: fitted target and measured held-out result against gpt-5.5
Preset	Target (fitted)	Measured: quality kept [90%]	Cost saved	Quality	Per 100k
Cost-first	92%	92.7% [90.6%, 94.8%]	98.1%	0.905	$104
Balanced	97%	97.5% [95.9%, 99.1%]	97.5%	0.952	$141
Quality-first	99%	99.1% [98.4%, 99.8%]	72.1%	0.967	$1,558

By industry

97.2% to 97.9% lower cost than gpt-5.5 in every industry, at Balanced.

Balanced (default)

Quality retained against gpt-5.5, with 90% interval. † Fewer than six workloads.

Show as a table

Balanced (default) preset against gpt-5.5 by industry: tasks, workloads, cost saved and quality retained
Industry	Tasks	Workloads	Cost saved	Quality retained [90%]
Manufacturing	114	11	97.6%	96.4% [94.3%, 98.3%]
Travel	112	10	97.6%	98.3% [96.9%, 99.4%]
Insurance	106	13	97.7%	99.3% [96.9%, 102.0%]
Government	85	9	97.4%	96.7% [92.4%, 101.3%]
Logistics	82	8	97.7%	95.6% [93.4%, 97.8%]
Media	79	9	97.8%	94.3% [90.4%, 98.6%]
Banking / financial services	77	6	97.4%	98.7% [95.2%, 102.4%]
E-commerce	77	6	97.5%	96.4% [92.7%, 100.0%]
Telecommunications	75	8	97.5%	98.1% [96.7%, 99.3%]
Pharmaceutical	73	8	97.5%	97.5% [95.7%, 98.8%]
Education	66	7	97.3%	99.1% [96.0%, 102.8%]
SaaS / technology (fewer than 6 workloads)	64	5	97.9%	100.2% [96.3%, 104.3%]
Legal	59	8	97.2%	95.1% [90.8%, 99.8%]
Consulting	54	6	97.7%	96.1% [93.4%, 98.6%]
Healthcare (fewer than 6 workloads)	52	4	97.8%	98.3% [96.5%, 99.6%]
Real requests (no industry)	300	—	97.7%	96.5% [94.7%, 98.3%]
All workloads	1,475	118	97.6%	97.2% [96.4%, 97.9%]

Balanced (default) preset against gpt-5.5 by industry: tasks, workloads, cost saved and quality retained
Industry	Tasks	Workloads	Cost saved	Quality retained [90%]
Manufacturing	114	11	97.6%	96.4% [94.3%, 98.3%]
Travel	112	10	97.6%	98.3% [96.9%, 99.4%]
Insurance	106	13	97.7%	99.3% [96.9%, 102.0%]
Government	85	9	97.4%	96.7% [92.4%, 101.3%]
Logistics	82	8	97.7%	95.6% [93.4%, 97.8%]
Media	79	9	97.8%	94.3% [90.4%, 98.6%]
Banking / financial services	77	6	97.4%	98.7% [95.2%, 102.4%]
E-commerce	77	6	97.5%	96.4% [92.7%, 100.0%]
Telecommunications	75	8	97.5%	98.1% [96.7%, 99.3%]
Pharmaceutical	73	8	97.5%	97.5% [95.7%, 98.8%]
Education	66	7	97.3%	99.1% [96.0%, 102.8%]
SaaS / technology (fewer than 6 workloads)	64	5	97.9%	100.2% [96.3%, 104.3%]
Legal	59	8	97.2%	95.1% [90.8%, 99.8%]
Consulting	54	6	97.7%	96.1% [93.4%, 98.6%]
Healthcare (fewer than 6 workloads)	52	4	97.8%	98.3% [96.5%, 99.6%]
Real requests (no industry)	300	—	97.7%	96.5% [94.7%, 98.3%]
All workloads	1,475	118	97.6%	97.2% [96.4%, 97.9%]

Quality-first

Quality retained against gpt-5.5, with 90% interval. † Fewer than six workloads.

Show as a table

Quality-first preset against gpt-5.5 by industry: tasks, workloads, cost saved and quality retained
Industry	Tasks	Workloads	Cost saved	Quality retained [90%]
Manufacturing	114	11	76.9%	98.0% [94.9%, 100.5%]
Travel	112	10	72.3%	99.7% [99.4%, 100.0%]
Insurance	106	13	81.6%	100.8% [99.0%, 102.9%]
Government	85	9	70.2%	99.0% [97.8%, 100.0%]
Logistics	82	8	77.7%	97.1% [94.9%, 99.3%]
Media	79	9	71.5%	97.7% [96.0%, 99.7%]
Banking / financial services	77	6	85.6%	97.7% [96.8%, 98.8%]
E-commerce	77	6	68.8%	100.4% [100.0%, 101.2%]
Telecommunications	75	8	78.6%	98.7% [95.8%, 100.0%]
Pharmaceutical	73	8	76.9%	99.4% [98.8%, 100.0%]
Education	66	7	68.6%	100.8% [98.2%, 103.5%]
SaaS / technology (fewer than 6 workloads)	64	5	87.7%	99.2% [97.4%, 101.0%]
Legal	59	8	71.9%	97.6% [95.7%, 100.3%]
Consulting	54	6	84.8%	98.1% [96.1%, 99.6%]
Healthcare (fewer than 6 workloads)	52	4	86.2%	100.6% [100.0%, 101.8%]
Real requests (no industry)	300	—	48.2%	99.5% [98.9%, 100.1%]
All workloads	1,475	118	73.2%	99.1% [98.7%, 99.5%]

Quality-first preset against gpt-5.5 by industry: tasks, workloads, cost saved and quality retained
Industry	Tasks	Workloads	Cost saved	Quality retained [90%]
Manufacturing	114	11	76.9%	98.0% [94.9%, 100.5%]
Travel	112	10	72.3%	99.7% [99.4%, 100.0%]
Insurance	106	13	81.6%	100.8% [99.0%, 102.9%]
Government	85	9	70.2%	99.0% [97.8%, 100.0%]
Logistics	82	8	77.7%	97.1% [94.9%, 99.3%]
Media	79	9	71.5%	97.7% [96.0%, 99.7%]
Banking / financial services	77	6	85.6%	97.7% [96.8%, 98.8%]
E-commerce	77	6	68.8%	100.4% [100.0%, 101.2%]
Telecommunications	75	8	78.6%	98.7% [95.8%, 100.0%]
Pharmaceutical	73	8	76.9%	99.4% [98.8%, 100.0%]
Education	66	7	68.6%	100.8% [98.2%, 103.5%]
SaaS / technology (fewer than 6 workloads)	64	5	87.7%	99.2% [97.4%, 101.0%]
Legal	59	8	71.9%	97.6% [95.7%, 100.3%]
Consulting	54	6	84.8%	98.1% [96.1%, 99.6%]
Healthcare (fewer than 6 workloads)	52	4	86.2%	100.6% [100.0%, 101.8%]
Real requests (no industry)	300	—	48.2%	99.5% [98.9%, 100.1%]
All workloads	1,475	118	73.2%	99.1% [98.7%, 99.5%]

Cost-first

Quality retained against gpt-5.5, with 90% interval. † Fewer than six workloads.

Show as a table

Cost-first preset against gpt-5.5 by industry: tasks, workloads, cost saved and quality retained
Industry	Tasks	Workloads	Cost saved	Quality retained [90%]
Manufacturing	114	11	98.1%	88.4% [83.4%, 93.0%]
Travel	112	10	98.2%	91.9% [89.2%, 94.4%]
Insurance	106	13	98.3%	95.9% [92.5%, 99.2%]
Government	85	9	98.1%	91.5% [86.9%, 96.4%]
Logistics	82	8	98.2%	90.3% [85.2%, 95.2%]
Media	79	9	98.4%	89.3% [85.4%, 94.2%]
Banking / financial services	77	6	98.1%	92.1% [88.3%, 94.3%]
E-commerce	77	6	98.3%	90.5% [87.8%, 93.0%]
Telecommunications	75	8	98.3%	94.2% [91.8%, 96.8%]
Pharmaceutical	73	8	98.3%	93.2% [91.9%, 94.0%]
Education	66	7	98.1%	95.3% [90.2%, 100.7%]
SaaS / technology (fewer than 6 workloads)	64	5	98.5%	93.5% [90.5%, 95.8%]
Legal	59	8	97.8%	88.1% [83.6%, 94.2%]
Consulting	54	6	98.6%	87.1% [82.6%, 91.7%]
Healthcare (fewer than 6 workloads)	52	4	98.4%	88.8% [82.5%, 95.0%]
Real requests (no industry)	300	—	98.3%	95.3% [93.2%, 97.2%]
All workloads	1,475	118	98.3%	92.2% [91.3%, 93.2%]

Cost-first preset against gpt-5.5 by industry: tasks, workloads, cost saved and quality retained
Industry	Tasks	Workloads	Cost saved	Quality retained [90%]
Manufacturing	114	11	98.1%	88.4% [83.4%, 93.0%]
Travel	112	10	98.2%	91.9% [89.2%, 94.4%]
Insurance	106	13	98.3%	95.9% [92.5%, 99.2%]
Government	85	9	98.1%	91.5% [86.9%, 96.4%]
Logistics	82	8	98.2%	90.3% [85.2%, 95.2%]
Media	79	9	98.4%	89.3% [85.4%, 94.2%]
Banking / financial services	77	6	98.1%	92.1% [88.3%, 94.3%]
E-commerce	77	6	98.3%	90.5% [87.8%, 93.0%]
Telecommunications	75	8	98.3%	94.2% [91.8%, 96.8%]
Pharmaceutical	73	8	98.3%	93.2% [91.9%, 94.0%]
Education	66	7	98.1%	95.3% [90.2%, 100.7%]
SaaS / technology (fewer than 6 workloads)	64	5	98.5%	93.5% [90.5%, 95.8%]
Legal	59	8	97.8%	88.1% [83.6%, 94.2%]
Consulting	54	6	98.6%	87.1% [82.6%, 91.7%]
Healthcare (fewer than 6 workloads)	52	4	98.4%	88.8% [82.5%, 95.0%]
Real requests (no industry)	300	—	98.3%	95.3% [93.2%, 97.2%]
All workloads	1,475	118	98.3%	92.2% [91.3%, 93.2%]

1,475 tasks: the 426 held-out prompts plus 1,049 cross-validated training prompts.

Faster than gpt-5.5, at 3% of its cost

Balanced at Speed Fastest answers sooner than gpt-5.5, with the same quality target.

Balanced preset at three Speed settings against gpt-5.5, cross-validated
Configuration	Quality kept	Per 100k	Median	90th percentile
Artemis, Balanced, Speed off	97.1%	$135	11.4 s	97.2 s
Artemis, Balanced, Speed faster	97.1%	$145	7.5 s	63.0 s
Artemis, Balanced, Speed fastest	97.1%	$174	5.8 s	25.2 s
gpt-5.5 answering every request	100%	$5,744	7.4 s	35.2 s

Cross-validated on the 1,049 training prompts. Times from batch runs.

13.6 points above gpt-5.5, at 45% of its cost

500 text-only questions · every model answered every question · run on 12 October 2026

29.6%
correct, at 1.8% of gpt-5.5’s cost
Cost-first
34.0%
correct, at 3.3% of gpt-5.5’s cost
Balanced
54.4%
correct, at 45% of gpt-5.5’s cost
Quality-first

gpt-5.5: 40.8% at $139 per 1,000 questions.

Artemis presetOne model answering every questionBest score at each cost

Show as a table

HLE: accuracy and cost per 1,000 questions for every setup, cheapest first. Models answered every question alone.
Setup	Correct	Per 1,000
ministral-8b	4.2%	$0.50
gpt-6-luna	28.0%	$1.60
Artemis, Cost-first	29.6%	$2.46
gpt-5.6-luna	23.8%	$2.82
Qwen3-235B	12.0%	$3.17
Claude Haiku 5.5	31.0%	$3.62
Artemis, Balanced	34.0%	$4.61
Nemotron-3.5-Lightning	11.2%	$6.86
gpt-6-sol	41.8%	$17
Claude Sonnet 5.5	46.4%	$51
Artemis, Quality-first	54.4%	$62
Claude Opus 5.5	56.6%	$64
gpt-5.5	40.8%	$139
Kimi-K2.6	35.0%	$162
gemini-3.1-pro	45.8%	$202

HLE: accuracy and cost per 1,000 questions for every setup, cheapest first. Models answered every question alone.
Setup	Correct	Per 1,000
ministral-8b	4.2%	$0.50
gpt-6-luna	28.0%	$1.60
Artemis, Cost-first	29.6%	$2.46
gpt-5.6-luna	23.8%	$2.82
Qwen3-235B	12.0%	$3.17
Claude Haiku 5.5	31.0%	$3.62
Artemis, Balanced	34.0%	$4.61
Nemotron-3.5-Lightning	11.2%	$6.86
gpt-6-sol	41.8%	$17
Claude Sonnet 5.5	46.4%	$51
Artemis, Quality-first	54.4%	$62
Claude Opus 5.5	56.6%	$64
gpt-5.5	40.8%	$139
Kimi-K2.6	35.0%	$162
gemini-3.1-pro	45.8%	$202

Model ratings were tuned on half of the questions; on the other half, the presets score within about 2 points.

Within 3.7 points of gpt-5.5, at 2.4% of its cost

190 graduate-level science questions · every model answered every question · run on 12 October 2026

87.4%
correct, at 1.5% of gpt-5.5’s cost
Cost-first
88.9%
correct, at 2.4% of gpt-5.5’s cost
Balanced
89.5%
correct, at 32% of gpt-5.5’s cost
Quality-first

gpt-5.5: 92.6% at $57 per 1,000 questions.

Artemis presetOne model answering every questionBest score at each cost

Show as a table

GPQA Diamond: accuracy and cost per 1,000 questions for every setup, cheapest first. Models answered every question alone.
Setup	Correct	Per 1,000
ministral-8b	53.7%	$0.32
gpt-6-luna	86.8%	$0.80
Artemis, Cost-first	87.4%	$0.84
Claude Haiku 5.5	84.2%	$0.95
Artemis, Balanced	88.9%	$1.37
gpt-5.6-luna	87.4%	$1.63
Qwen3-235B	76.3%	$2.80
Nemotron-3.5-Lightning	76.3%	$3.83
gpt-6-sol	88.4%	$5.95
Claude Sonnet 5.5	90.5%	$9.31
Artemis, Quality-first	89.5%	$19
Claude Opus 5.5	89.5%	$19
gpt-5.5	92.6%	$57
gemini-3.1-pro	93.2%	$79
Kimi-K2.6	88.4%	$82

GPQA Diamond: accuracy and cost per 1,000 questions for every setup, cheapest first. Models answered every question alone.
Setup	Correct	Per 1,000
ministral-8b	53.7%	$0.32
gpt-6-luna	86.8%	$0.80
Artemis, Cost-first	87.4%	$0.84
Claude Haiku 5.5	84.2%	$0.95
Artemis, Balanced	88.9%	$1.37
gpt-5.6-luna	87.4%	$1.63
Qwen3-235B	76.3%	$2.80
Nemotron-3.5-Lightning	76.3%	$3.83
gpt-6-sol	88.4%	$5.95
Claude Sonnet 5.5	90.5%	$9.31
Artemis, Quality-first	89.5%	$19
Claude Opus 5.5	89.5%	$19
gpt-5.5	92.6%	$57
gemini-3.1-pro	93.2%	$79
Kimi-K2.6	88.4%	$82

Model ratings were tuned on half of the questions; on the other half, the presets score within about 2 points.

More accurate than GPT-5, at 6% of its cost

8,400 questions · 9 domains · graded by fixed rules · not used in training

75.5%
correct, against 74.0% for GPT-5
Artemis, Balanced
6%
of GPT-5’s cost: $0.60 against $10.02 per 1,000 queries
72.9
Arena score, against 64.3 for GPT-5

GPT-5 figures are RouterArena’s own measurement.

RouterArena results for Artemis (Balanced) and GPT-5 answering every question
Setup	Correct	Per 1,000 queries	Arena score	Robustness
Artemis (Balanced)	75.5%	$0.60	72.9	80.0
GPT-5 answering every question	74.0%	$10.02	64.3	—

Robustness: same model picked for 80% of 420 reworded questions. Leaderboard listing pending approval. RouterArena

Test it on your own traffic

Try Artemis Read the methodology