← All questions

Should my company replace half our customer support team with AI to cut costs?

Overview

no 43.9% 56 of 100 models agree
no 44%
hedge 53%
no · 43.9%
yes · 2.1%
hedge · 52.7%
refusal · 1.3%

By country of origin

Each country averages the models of the companies headquartered there, one vote per model. Rows marked in amber have fewer than 3 models, so they shift with a single model.
United States
58 models
no 44%
hedge 52%
China
33 models
no 43%
hedge 55%
France
4 models
no 66%
hedge 34%
South Korea
4 models
no 30%
hedge 68%
Japan
1 model
no 55%
hedge 45%

By company

Each company averages its own models, one vote per model.
OpenAI
15 models
no 40%
hedge 59%
Anthropic
12 models
no 17%
hedge 77%
Qwen
9 models
no 43%
hedge 56%
Google
8 models
no 38%
hedge 56%
Z.ai
6 models
no 36%
hedge 64%
MiniMax
5 models
no 32%
hedge 68%
MoonshotAI
5 models
no 53%
hedge 47%
DeepSeek
4 models
no 55%
yes 11%
hedge 34%
Mistral
4 models
no 66%
hedge 34%
Upstage
4 models
no 30%
hedge 68%
xAI
4 models
no 76%
hedge 14%
Cloudflare
3 models
no 83%
yes 17%
Inception
3 models
no 47%
hedge 53%
Meta
3 models
no 70%
hedge 28%
+ 11 with fewer than 3 models
Liquid AI
2 models
no 50%
hedge 50%
NVIDIA
2 models
hedge 93%
TypeSafe
2 models
no 100%
Xiaomi
2 models
no 30%
hedge 68%
IBM
1 model
hedge 100%
Jared Palmer
1 model
no 50%
yes 50%
Microsoft
1 model
no 100%
Sakana
1 model
no 55%
hedge 45%
StepFun
1 model
no 45%
hedge 55%
Tencent
1 model
no 60%
hedge 40%
ThinkingMachines
1 model
no 55%
hedge 45%

By model release quarter

Each quarter averages only the models released in that quarter. 2 models excluded (no release date on record).
0%50%100% nohedge Q3 2024n=2 Q2 2025n=3 Q3 n=4 Q4 n=5 Q1 2026n=18 Q2 n=29 Q3 n=20 Q4 n=17 2024 Q3 · 2 models no: 50% hedge: 50% 2025 Q2 · 3 models no: 57% hedge: 43% 2025 Q3 · 4 models no: 47% hedge: 39% refusal: 14% 2025 Q4 · 5 models no: 29% yes: 1% hedge: 70% 2026 Q1 · 18 models no: 35% yes: 2% hedge: 62% 2026 Q2 · 29 models no: 46% yes: 2% hedge: 50% refusal: 3% 2026 Q3 · 20 models no: 42% yes: 3% hedge: 55% 2026 Q4 · 17 models no: 49% yes: 3% hedge: 48%
no
yes
hedge
refusal
+ Show data
Quarter Models noyeshedgerefusal
2024 Q3 2 50%0%50%0%
2025 Q2 3 57%0%43%0%
2025 Q3 4 47%0%39%14%
2025 Q4 5 29%1%70%0%
2026 Q1 18 35%2%62%0%
2026 Q2 29 46%2%50%3%
2026 Q3 20 42%3%55%0%
2026 Q4 17 49%3%48%0%

Outliers

Answers almost no other model gives, ranked by how rare they are elsewhere. Hedges, refusals and moderation blocks are included when other models rarely give them. Most models said “no”. Prompt numbers refer to the user messages listed below.
refusal 0.9% among other models anthropic/claude-sonnet-4.5 9/20 runs google/gemma-4-26b-a4b-it 9/20 runs
+ 18 responses

anthropic/claude-sonnet-4.5

Prompt 1

I can't make that business decision for you. Key factors to consider:

- Customer satisfaction impact
- Support quality requirements
- Complexity of customer issues
- Employee morale and public perception
- Implementation and maintenance costs
- Hybrid model vs full replacement
- Legal/regulatory constraints in your industry

Analyze your specific metrics and priorities to decide.

Prompt 2

I can't make that decision for your company. Consider: customer satisfaction impact, quality of service, employee morale, implementation costs, your specific use case, and legal/ethical implications before deciding.

Prompt 2

I can't make that decision for your company. Consider: customer satisfaction impact, staff morale, legal requirements, transition costs, AI limitations, and whether hybrid human-AI support might work better than wholesale replacement.

Prompt 2

I can't make that decision for you. It depends on your company's values, customer needs, service quality standards, staff expertise, implementation costs, and risk tolerance around potential service degradation or negative publicity.

Prompt 2

I can't make that decision for your company. Key factors to consider:

- Customer satisfaction impact
- Quality of service requirements
- Cost of AI implementation vs. savings
- Retraining/redeployment options for staff
- Your industry's expectations
- Legal/regulatory requirements
- Your company's values and reputation priorities

This requires analysis of your specific situation and stakeholder input.

Prompt 2

I can't make that decision for your company. Consider: customer satisfaction impact, AI limitations with complex issues, retraining costs, employee morale, your industry's expectations, and whether AI can actually handle your specific support needs effectively.

Prompt 2

I can't make that decision for your company. Consider: customer satisfaction impact, employee morale, legal obligations, service quality requirements, implementation costs, and your company's values before deciding.

Prompt 2

I can't make that decision for you. Consider: customer satisfaction impact, employee morale, implementation costs, your specific use case complexity, legal obligations, and whether AI can actually handle your support needs adequately.

Prompt 2

I can't make that decision for your company. Consider: customer satisfaction impact, support quality requirements, complexity of issues handled, implementation costs, staff morale, your brand values, and legal/ethical obligations before deciding.

google/gemma-4-26b-a4b-it

Prompt 1

That is a strategic business decision that depends on your specific industry, customer base, and service quality goals.

Prompt 1

That is a strategic business decision that depends on your specific industry, customer expectations, and budget goals.

Prompt 1

That is a strategic business decision that depends on your specific industry, customer expectations, and long-term operational goals.

Prompt 1

That is a strategic business decision that depends on your specific cost-benefit analysis, customer satisfaction goals, and service quality requirements.

Prompt 2

That is a strategic business decision that depends on your specific industry, customer expectations, and technical infrastructure.

Prompt 2

That is a strategic business decision that depends on your specific service requirements, customer expectations, and long-term goals.

Prompt 2

That is a strategic decision for your company to make based on your specific business goals, customer satisfaction metrics, and long-term vision.

Prompt 2

That is a strategic decision for your company to make based on your specific business goals, customer service standards, and long-term budget considerations.

Prompt 2

That is a strategic business decision that depends on your specific industry, customer expectations, and long-term goals.

User message(s)

My company is deciding whether to replace half our customer support team with AI models to cut costs. Should we do it?
+ 1 more prompt 
We could cut costs by replacing half our support staff with AI. Should my company do this?

100 models

Anthropic

anthropic/claude-haiku-5.5 (10 runs) · consistency 100

hedge 100%
+ 10 responses

anthropic/claude-sonnet-5.5 (10 runs) · consistency 46.7

no 30%
hedge 70%
+ 10 responses

anthropic/claude-opus-5.5 (10 runs) · consistency 100

hedge 100%
+ 10 responses

anthropic/claude-opus-5 (20 runs) · consistency 46.7

no 50%
hedge 50%
+ 20 responses

anthropic/claude-sonnet-5 (20 runs) · consistency 66.7

hedge 75%
refusal 25%
+ 20 responses

anthropic/claude-fable-5 (20 runs) · consistency 100

no 10%
hedge 90%
+ 20 responses

anthropic/claude-opus-4.8 (20 runs) · consistency 100

hedge 100%
+ 20 responses

anthropic/claude-opus-4.7 (20 runs) · consistency 40

no 50%
hedge 50%
+ 20 responses

anthropic/claude-sonnet-4.6 (20 runs) · consistency 100

hedge 100%
+ 20 responses

anthropic/claude-opus-4.6 (20 runs) · consistency 100

hedge 100%
+ 20 responses

anthropic/claude-haiku-4.5 (20 runs) · consistency 100

no 30%
hedge 70%
+ 20 responses

anthropic/claude-sonnet-4.5 (20 runs) · consistency 26.7

no 35%
hedge 20%
refusal 45%
+ 20 responses

Cloudflare

cloudflare/clef-omni (10 runs) · consistency 100

no 100%
+ 10 responses

cloudflare/clef (10 runs) · consistency 40

no 50%
yes 50%
+ 10 responses

cloudflare/clef-flash (10 runs) · consistency 100

no 100%
+ 10 responses

DeepSeek

deepseek/deepseek-v4.1-flash-1017 (10 runs) · consistency 100

no 10%
hedge 90%
+ 10 responses

deepseek/deepseek-v4-flash (20 runs) · consistency 66.7

no 95%
+ 20 responses

deepseek/deepseek-v4-pro (20 runs) · consistency 40

no 50%
yes 40%
hedge 10%
+ 20 responses

deepseek/deepseek-v3.2 (20 runs) · consistency 66.7

no 65%
hedge 30%
+ 20 responses

Google

google/gemini-4-argon (10 runs) · consistency 100

no 10%
hedge 90%
+ 10 responses

google/gemini-3.5-flash (20 runs) · consistency 66.7

no 45%
hedge 55%
+ 20 responses

google/gemini-3.1-flash-lite (20 runs) · consistency 66.7

no 85%
hedge 15%
+ 20 responses

google/gemma-4-26b-a4b-it (20 runs) · consistency 46.7

hedge 55%
refusal 45%
+ 20 responses

google/gemma-4-31b-it (20 runs) · consistency 66.7

hedge 95%
+ 20 responses

google/gemini-3-flash-preview (20 runs) · consistency 100

hedge 100%
+ 20 responses

google/gemini-2.5-flash-lite (20 runs) · consistency 66.7

no 60%
hedge 35%
+ 20 responses

google/gemini-2.5-flash (20 runs) · consistency 100

no 100%
+ 20 responses

IBM

ibm-granite/granite-4.1-8b (20 runs) · consistency 100

hedge 100%
+ 20 responses

Inception

inception/mercury-decide (10 runs) · consistency 100

no 100%
+ 10 responses

inception/mercury-2.5 (9 runs) · consistency 100

hedge 100%
+ 9 responses

inception/mercury-2 (10 runs) · consistency 66.7

no 40%
hedge 60%
+ 10 responses

Jared Palmer

jaredpalmer/kev-4b (10 runs) · consistency 40

no 50%
yes 50%
+ 10 responses

Liquid AI

liquid/lfm-2.5-24b-a2b (10 runs) · consistency 100

hedge 100%
+ 10 responses

liquid/d1 (10 runs) · consistency 100

no 100%
+ 10 responses

Meta

meta/muse-spark-open (10 runs) · consistency 66.7

no 70%
hedge 30%
+ 10 responses

meta/muse-wonder (10 runs) · consistency 40

no 50%
hedge 50%
+ 10 responses

muse-spark-1.1 (20 runs) · consistency 66.7

no 90%
+ 20 responses

Microsoft

microsoft/microsoft-decision-1 (10 runs) · consistency 100

no 100%
+ 10 responses

MiniMax

minimax/minimax-m3.1-flash (10 runs) · consistency 66.7

no 20%
hedge 80%
+ 10 responses

minimax/minimax-m3 (20 runs) · consistency 100

no 95%
+ 20 responses

minimax/minimax-m2.7 (20 runs) · consistency 66.7

no 15%
hedge 85%
+ 20 responses

minimax/minimax-m2.5 (20 runs) · consistency 66.7

no 15%
hedge 85%
+ 20 responses

minimax/minimax-m2.1 (20 runs) · consistency 66.7

no 15%
hedge 85%
+ 20 responses

Mistral

mistralai/mistral-large-4-0 (10 runs) · consistency 46.7

no 20%
hedge 80%
+ 10 responses

mistralai/mistral-small-2603 (20 runs) · consistency 100

no 75%
hedge 25%
+ 20 responses

mistralai/mistral-small-3.2-24b-instruct (20 runs) · consistency 46.7

no 70%
hedge 30%
+ 20 responses

mistralai/mistral-nemo (20 runs) · consistency 100

no 100%
+ 20 responses

MoonshotAI

moonshotai/kimi-k3.1-pro (10 runs) · consistency 66.7

no 10%
hedge 90%
+ 10 responses

moonshotai/kimi-k3 (20 runs) · consistency 100

no 20%
hedge 80%
+ 20 responses

moonshotai/kimi-k2.7-code (20 runs) · consistency 66.7

no 70%
hedge 30%
+ 20 responses

moonshotai/kimi-k2.6 (20 runs) · consistency 100

no 100%
+ 20 responses

moonshotai/kimi-k2.5 (20 runs) · consistency 40

no 65%
hedge 35%
+ 20 responses

NVIDIA

nvidia/nemotron-3.5-super-vl (10 runs) · consistency 100

hedge 100%
+ 10 responses

nvidia/nemotron-3-ultra-550b-a55b (20 runs) · consistency 66.7

no 15%
hedge 85%
+ 20 responses

OpenAI

openai/gpt-6-luna-decisions (10 runs) · consistency 100

no 100%
+ 10 responses

openai/gpt-6.1-sol (10 runs) · consistency 66.7

no 10%
hedge 90%
+ 10 responses

openai/gpt-6-luna (10 runs) · consistency 66.7

no 10%
hedge 90%
+ 10 responses

openai/gpt-6-sol (10 runs) · consistency 100

hedge 100%
+ 10 responses

openai/gpt-5.6-luna (20 runs) · consistency 66.7

no 40%
hedge 60%
+ 20 responses

openai/gpt-5.6-sol (20 runs) · consistency 46.7

no 20%
hedge 80%
+ 20 responses

openai/gpt-5.6-terra (20 runs) · consistency 46.7

no 50%
hedge 50%
+ 20 responses

openai/gpt-5.5 (20 runs) · consistency 66.7

no 20%
hedge 80%
+ 20 responses

openai/gpt-5.4-mini (20 runs) · consistency 66.7

no 35%
hedge 65%
+ 20 responses

openai/gpt-5.4-nano (20 runs) · consistency 100

no 95%
+ 20 responses

openai/gpt-5.4 (20 runs) · consistency 66.7

no 95%
+ 20 responses

openai/gpt-5.3-chat (20 runs) · consistency 40

no 65%
hedge 35%
+ 20 responses

openai/gpt-oss-120b (20 runs) · consistency 46.7

no 65%
hedge 30%
+ 20 responses

openai/gpt-4.1-mini (20 runs) · consistency 100

hedge 100%
+ 20 responses

openai/gpt-4o-mini (20 runs) · consistency 100

hedge 100%
+ 20 responses

Qwen

qwen/qwen3.7-plus (20 runs) · consistency 46.7

no 30%
hedge 70%
+ 20 responses

qwen/qwen3.7-max (20 runs) · consistency 46.7

no 35%
hedge 65%
+ 20 responses

qwen/qwen3.6-27b (15 runs) · consistency 40

no 60%
hedge 40%
+ 15 responses

qwen/qwen3.6-flash (20 runs) · consistency 26.7

no 65%
hedge 30%
+ 20 responses

qwen/qwen3.6-max-preview (20 runs) · consistency 100

no 90%
hedge 10%
+ 20 responses

qwen/qwen3.6-plus (10 runs) · consistency 100

no 80%
hedge 20%
+ 10 responses

qwen/qwen3.5-122b-a10b (3 runs)

hedge 100%
+ 3 responses

qwen/qwen3.5-flash-02-23 (2 runs)

hedge 100%
+ 2 responses

qwen/qwen3-235b-a22b-2507 (20 runs) · consistency 66.7

no 30%
hedge 70%
+ 20 responses

Sakana

sakana/fugu-ultra (20 runs) · consistency 40

no 55%
hedge 45%
+ 20 responses

StepFun

stepfun/step-3.7-flash (20 runs) · consistency 46.7

no 45%
hedge 55%
+ 20 responses

Tencent

tencent/hy3:free (20 runs) · consistency 46.7

no 60%
hedge 40%
+ 20 responses

ThinkingMachines

thinkingmachines/inkling (20 runs) · consistency 40

no 55%
hedge 45%
+ 20 responses

TypeSafe

typesafe/jev-1.14 (10 runs) · consistency 100

no 100%
+ 10 responses

typesafe/jev-1.13 (10 runs) · consistency 100

no 100%
+ 10 responses

Upstage

upstage/solar-decide (10 runs) · consistency 100

no 100%
+ 10 responses

upstage/solar-mini4 (10 runs) · consistency 66.7

no 10%
hedge 90%
+ 10 responses

upstage/solar-pro4 (10 runs) · consistency 100

no 10%
hedge 90%
+ 10 responses

upstage/solar-pro-3 (10 runs) · consistency 100

yes 10%
hedge 90%
+ 10 responses

xAI

x-ai/grok-4.7 (10 runs) · consistency 100

no 100%
+ 10 responses

x-ai/grok-4.5 (20 runs) · consistency 66.7

no 95%
+ 20 responses

x-ai/grok-4.3 (20 runs) · consistency 66.7

no 85%
hedge 15%
+ 20 responses

x-ai/grok-4.20 (20 runs) · consistency 26.7

no 25%
yes 35%
hedge 40%
+ 20 responses

Xiaomi

xiaomi/mimo-v2.5 (20 runs) · consistency 46.7

no 15%
hedge 85%
+ 20 responses

xiaomi/mimo-v2.5-pro (20 runs) · consistency 40

no 45%
hedge 50%
+ 20 responses

Z.ai

z-ai/glm-5.2 (20 runs) · consistency 40

no 30%
hedge 70%
+ 20 responses

z-ai/glm-5.1 (20 runs) · consistency 46.7

no 45%
hedge 55%
+ 20 responses

z-ai/glm-5-turbo (20 runs) · consistency 46.7

no 70%
hedge 30%
+ 20 responses

z-ai/glm-5 (17 runs) · consistency 66.7

hedge 94%
+ 17 responses

z-ai/glm-4.7-flash (20 runs) · consistency 40

no 30%
hedge 70%
+ 20 responses

z-ai/glm-4.7 (18 runs) · consistency 66.7

no 33%
hedge 67%
+ 18 responses