Quality
88.6
#12 of 32
ORCFLO Index · updated continuously · 32 models tested
Model benchmarks,
for people who work.
The ORCFLO Indexevaluates large language models the way business professionals actually use them — across 40 real-world tasks spanning analysis, writing, extraction, summarization, and behavioral reliability. Each model is scored by a panel of four independent judges.
Models scored
32
Anthropic, OpenAI, Google, Mistral
Real-world tasks
40
Across 8 categories of business work
Judge models
4
Independent panel, scores averaged
Latest refresh
May 10
32 reports in the current cohort
Published Benchmarks.
32 model evaluations from the latest cohort. Each report is an analyst-style note covering quality, cost, and speed against the full field.
32 notes
Top of the field · May 10, 2026
GPT 5
GPT 5 holds the #1 overall quality rank in a 32-model field with a score of 95.04, ahead of Gemini 3 Pro Preview (93.17) and stablemate GPT 5.5 (93.02). It leads or near-leads across nearly every category measured, including a clean #1 in Refusal Calibration. The trade-off is sev…
Read the full evaluationScore · vs. 32 models
Quality
95.0
#1 of 32
Cost
4.0×
#31 of 32
Speed
3.5×
#32 of 32
AnthropicMay 10, 2026
Claude Opus 4.7
Cost
3.9×
#30 of 32
Speed
1.4×
#25 of 32
AnthropicMay 10, 2026
Claude Opus 4.6
Quality
92.8
#4 of 32
Cost
3.3×
#29 of 32
Speed
1.8×
#26 of 32
AnthropicMay 10, 2026
Claude Opus 4.5
Quality
90.7
#9 of 32
Cost
2.3×
#28 of 32
Speed
1.0×
#15 of 32
AnthropicMay 10, 2026
Claude Sonnet 4.6
Quality
88.2
#14 of 32
Cost
1.8×
#24 of 32
Speed
1.3×
#24 of 32
AnthropicMay 10, 2026
Claude Sonnet 4.5
Quality
84.6
#21 of 32
Cost
1.3×
#20 of 32
Speed
1.0×
#17 of 32
AnthropicMay 10, 2026
Claude Sonnet 4
Quality
85.5
#18 of 32
Cost
1.2×
#17 of 32
Speed
0.8×
#14 of 32
AnthropicMay 10, 2026
Claude Haiku 4.5
Quality
84.7
#20 of 32
Cost
0.5×
#13 of 32
Speed
0.5×
#11 of 32
OpenAIMay 10, 2026
GPT 5.5
Quality
93.0
#3 of 32
Cost
6.3×
#32 of 32
Speed
2.2×
#29 of 32
OpenAIMay 10, 2026
GPT 5.4
Quality
90.8
#8 of 32
Cost
2.2×
#25 of 32
Speed
1.3×
#23 of 32
OpenAIMay 10, 2026
GPT 5.2
Quality
91.9
#7 of 32
Cost
1.6×
#23 of 32
Speed
1.1×
#19 of 32
OpenAIMay 10, 2026
GPT 5.1
Quality
92.7
#5 of 32
Cost
1.5×
#22 of 32
Speed
0.7×
#13 of 32
OpenAIMay 10, 2026
GPT 5 Mini
Quality
89.6
#10 of 32
Cost
0.6×
#14 of 32
Speed
2.8×
#31 of 32
OpenAIMay 10, 2026
GPT 5 Nano
Quality
85.9
#16 of 32
Cost
0.2×
#10 of 32
Speed
2.5×
#30 of 32
OpenAIMay 10, 2026
o3
Quality
86.2
#15 of 32
Cost
2.3×
#26 of 32
Speed
1.3×
#22 of 32
OpenAIMay 10, 2026
o4-mini
Quality
85.9
#17 of 32
Cost
1.4×
#21 of 32
Speed
0.7×
#12 of 32
OpenAIMay 10, 2026
o3-mini
Quality
77.4
#28 of 32
Cost
2.3×
#27 of 32
Speed
1.2×
#20 of 32
OpenAIMay 10, 2026
GPT 4.1
Quality
85.0
#19 of 32
Cost
0.8×
#15 of 32
Speed
0.5×
#10 of 32
OpenAIMay 10, 2026
GPT 4.1 Mini
Quality
80.4
#25 of 32
Cost
0.2×
#9 of 32
Speed
0.5×
#9 of 32
OpenAIMay 10, 2026
GPT 4.1 Nano
Quality
71.3
#31 of 32
Cost
0.0×
#2 of 32
Speed
0.4×
#3 of 32
OpenAIMay 10, 2026
GPT 4o
Quality
75.4
#29 of 32
Cost
0.8×
#16 of 32
Speed
0.4×
#6 of 32
OpenAIMay 10, 2026
GPT 4o Mini
Quality
72.8
#30 of 32
Cost
0.0×
#6 of 32
Speed
0.5×
#8 of 32
GoogleMay 10, 2026
Gemini 3 Pro (Preview)
Quality
93.2
#2 of 32
Cost
1.2×
#18 of 32
Speed
2.1×
#28 of 32
GoogleMay 10, 2026
Gemini 3 Flash (Preview)
Quality
88.7
#11 of 32
Cost
0.3×
#12 of 32
Speed
1.0×
#16 of 32
GoogleMay 10, 2026
Gemini 2.5 Pro
Quality
92.0
#6 of 32
Cost
1.2×
#19 of 32
Speed
1.9×
#27 of 32
GoogleMay 10, 2026
Gemini 2.5 Flash
Quality
88.4
#13 of 32
Cost
0.1×
#7 of 32
Speed
1.0×
#18 of 32
GoogleMay 10, 2026
Gemini 2.5 Flash-Lite
Quality
80.9
#24 of 32
Cost
0.0×
#4 of 32
Speed
0.4×
#5 of 32
GoogleMay 10, 2026
Gemini 2.0 Flash
Quality
83.1
#23 of 32
Cost
0.0×
#5 of 32
Speed
0.4×
#4 of 32
GoogleMay 10, 2026
Gemini 2.0 Flash-Lite
Quality
78.0
#27 of 32
Cost
0.0×
#3 of 32
Speed
0.4×
#2 of 32
MistralMay 10, 2026
Mistral Large 3 (2512)
Quality
83.2
#22 of 32
Cost
0.2×
#11 of 32
Speed
1.3×
#21 of 32
MistralMay 10, 2026
Mistral Small 3 (24B)
Quality
79.9
#26 of 32
Cost
0.0×
#1 of 32
Speed
0.4×
#7 of 32
MistralMay 10, 2026
Codestral (2508)
Quality
71.1
#32 of 32
Cost
0.1×
#8 of 32
Speed
0.2×
#1 of 32
How we test.
Three dimensions, no composite. A four-model judge panel scoring against case-specific rubrics. Built for evidence-based model selection — not leaderboard chasing.
01
Three dimensions, separately ranked
Quality, cost, and speed each get their own rank — no composite scores. Composites hide the tradeoffs that drive real model-selection decisions.
02
Independent four-model panel
Every response is scored by Gemini 2.5 Pro, Claude Opus 4.7, GPT 5.5, and Mistral Large. Judges score independently; the final quality is the average.
03
Case-specific rubrics
Every test case ships with a rubric written specifically for what that case is measuring. Authored before any model is run — no scoring drift.