Best LLMs for Customer Support Chatbots
Short conversational turns where latency and cost per conversation matter.
Typical request this page prices
- Quality floor: Top 50% of priced models by Intelligence Index (≥ 14.9).
- Costs use each provider’s current standard rates, with cached tokens at the cached-input rate where one is published. Speed is from Artificial Analysis where measured.
Lowest cost that meets the bar
Models that clear the quality floor, ordered by estimated cost for the request above.
| Model | Intelligence | Speed | Cheapest host | Input / output per 1M | Per 1,000 requests |
|---|---|---|---|---|---|
| Ling-3.0-flash-VL InclusionAI | 24.6 | 147tok/s | OpenRouter 3 providers | $0.021 / $0.062 cached $0.0042 | $0.062 |
| Ling 3.0 Flash InclusionAI | 20.1 | - | OpenRouter 3 providers | $0.021 / $0.063 cached $0.0042 | $0.063 |
| Ling-3.0-flash-Fin InclusionAI | 22.6 | 317tok/s | OpenRouter 3 providers | $0.042 / $0.123 cached $0.0084 | $0.125 |
| Gemma 4 26B Google · 262K context | 16.7 | 101tok/s | OpenRouter 7 providers | $0.042 / $0.220 cached $0.021 | $0.182 |
| DeepSeek V4 Flash DeepSeek · 1.0M context | 34.3 | - | Deep Infra 19 providers | $0.060 / $0.180 cached $0.015 | $0.185 |
| GLM-5.3-Flash Z AI | 41.8 | 59tok/s | Pareto Inference 17 providers | $0.060 / $0.200 cached $0.012 | $0.188 |
| MiMo-V2.5 Xiaomi | 25.2 | - | GMI Cloud 7 providers | $0.119 / $0.238 cached $0.0024 | $0.277 |
| GLM 4.7 Flash Zhipu | 14.9 | - | Novita AI 5 providers | $0.070 / $0.400 cached $0.010 | $0.280 |
Highest-scoring options
Ordered by Artificial Analysis Intelligence Index, with the cost of the same request.
| Model | Intelligence | Speed | Cheapest host | Input / output per 1M | Per 1,000 requests |
|---|---|---|---|---|---|
| Claude Opus 5.5 Anthropic | 57.6 | 97tok/s | Deep Infra 6 providers | $4.00 / $20.00 cached $0.200 | $14.30 |
| Claude Sonnet 5.5 Anthropic | 56 | 135tok/s | Anthropic 6 providers | $2.00 / $10.00 cached $0.100 | $7.15 |
| Claude Fable 5.1 Anthropic | 53.4 | 72tok/s | DigitalOcean 4 providers | $10.00 / $50.00 cached $0.250 | $35.38 |
| GPT-6 Astra OpenAI | 52.7 | 63tok/s | Velokey 7 providers | $3.00 / $15.00 cached $0.300 | $10.95 |
| GPT-6.1 Sol OpenAI | 51.8 | 65tok/s | Microsoft Azure 6 providers | $2.00 / $10.00 cached $0.100 | $7.15 |
| Claude Opus 5 Anthropic | 50.8 | - | Velokey 7 providers | $2.00 / $10.00 cached $0.200 | $7.30 |
| Claude Fable 5 Anthropic · 1.0M context | 49.6 | - | Velokey 6 providers | $4.00 / $20.00 cached $0.400 | $14.60 |
| Muse Spark 1.3 Meta | 48.1 | 115tok/s | OpenRouter 2 providers | $1.25 / $4.25 cached $0.150 | $3.80 |
Compare every model and provider on the LLM API pricing table.
How to choose an LLM for customer support chatbots
A support chatbot sends a fixed system prompt with policies and tone guidance, a few retrieved help-centre passages, and the recent turns of the conversation. Responses are short. The workload is high-volume and latency-sensitive: users notice a pause before the first word more than they notice a slightly better answer.
Because the system prompt is identical across conversations, a large share of input can be cached. The cost estimate here assumes half the input is a cached prefix, which is typical once the policy text and tool definitions are stable.
Mid-range models are usually sufficient for answering from provided documentation, so the quality floor sits at the median of the priced set. Compare time to first token and output speed alongside cost; both are shown where benchmark data exists.
Frequently Asked Questions
Do I need a top-tier model for a support chatbot?
Usually not. When answers come from retrieved documentation, the model mostly needs to follow instructions and stay on topic. Escalation logic and retrieval quality tend to matter more than the model’s general reasoning score.
What should I compare besides price?
Time to first token and output tokens per second, which govern how responsive the chat feels, and whether the provider supports prompt caching for your stable system prompt. Speed figures on this page come from Artificial Analysis where available.
How do I turn the per-request figure into a cost per conversation?
Multiply by the average number of assistant turns per conversation. Later turns carry more history, so if conversations run long, raise the input assumption or summarise older turns.