Models are sorted by estimated cost for 1 million input tokens and 500 thousand output tokens. Use this page when your first constraint is API spend.
Cheap LLM API Models for Chatbots
Chatbot workloads can become output-token heavy, so this guide highlights models with low standardized workload costs, clear output pricing, and practical alternatives.
50Models listed
1M + 500KCost example tokens
USD / 1MNormalized prices
Quick shortlist
Start with Ring-2.6-1T (free).
This guide is sorted by standard workload cost, so the first rows are the strongest budget shortlist before model-quality testing.
The ranking is a discovery aid, not a final recommendation. Always compare the model against your workload and verify provider pricing before production use.
Estimate your workload cost
Customize guide costs
This estimate uses normalized public API pricing per 1M tokens. It is a planning aid, not a billing quote. Verify provider pricing, limits, and terms before production use.
Model Ranking
Browse all models| Model | Provider | Prompt | Output | Example Cost | Your Cost | Context | Rank | Release |
|---|---|---|---|---|---|---|---|---|
 🔥Ring-2.6-1T (free) | inclusionAI | $0 | $0 | $0 | $0 | 262.14K | #10 | 2026-05-08 |
| 🔥Nemotron 3 Super (free) | NVIDIA | $0 | $0 | $0 | $0 | 262.14K | #12 | 2026-03-11 |
| 🔥Owl Alpha | OpenRouter | $0 | $0 | $0 | $0 | 1.05M | #17 | 2026-04-28 |
| CoBuddy (free) | Baidu Qianfan | $0 | $0 | $0 | $0 | 131.07K | Unranked | 2026-05-06 |
| Nemotron 3 Nano Omni (free) | NVIDIA | $0 | $0 | $0 | $0 | 256K | Unranked | 2026-04-28 |
| Laguna XS.2 (free) | Poolside | $0 | $0 | $0 | $0 | 131.07K | Unranked | 2026-04-28 |
| Laguna M.1 (free) | Poolside | $0 | $0 | $0 | $0 | 131.07K | Unranked | 2026-04-28 |
| Qianfan-OCR-Fast (free) | Baidu | $0 | $0 | $0 | $0 | 65.54K | Unranked | 2026-04-20 |
| Gemma 4 26B A4B (free) | $0 | $0 | $0 | $0 | 262.14K | Unranked | 2026-04-03 | |
| Gemma 4 31B (free) | $0 | $0 | $0 | $0 | 262.14K | Unranked | 2026-04-02 | |
| Trinity Large Thinking (free) | Arcee AI | $0 | $0 | $0 | $0 | 262.14K | Unranked | 2026-04-01 |
| Lyria 3 Pro Preview | $0 | $0 | $0 | $0 | 1.05M | Unranked | 2026-03-30 | |
| Lyria 3 Clip Preview | $0 | $0 | $0 | $0 | 1.05M | Unranked | 2026-03-30 | |
| MiniMax M2.5 (free) | MiniMax | $0 | $0 | $0 | $0 | 196.61K | Unranked | 2026-02-12 |
| Free Models Router | OpenRouter | $0 | $0 | $0 | $0 | 200K | Unranked | 2026-02-01 |
| LFM2.5-1.2B-Thinking (free) | LiquidAI | $0 | $0 | $0 | $0 | 32.77K | Unranked | 2026-01-20 |
| LFM2.5-1.2B-Instruct (free) | LiquidAI | $0 | $0 | $0 | $0 | 32.77K | Unranked | 2026-01-20 |
| Nemotron 3 Nano 30B A3B (free) | NVIDIA | $0 | $0 | $0 | $0 | 256K | Unranked | 2025-12-14 |
| Nemotron Nano 12B 2 VL (free) | NVIDIA | $0 | $0 | $0 | $0 | 128K | Unranked | 2025-10-28 |
| Qwen3 Next 80B A3B Instruct (free) | Qwen | $0 | $0 | $0 | $0 | 262.14K | Unranked | 2025-09-11 |
| Nemotron Nano 9B V2 (free) | NVIDIA | $0 | $0 | $0 | $0 | 128K | Unranked | 2025-09-05 |
| gpt-oss-120b (free) | OpenAI | $0 | $0 | $0 | $0 | 131.07K | Unranked | 2025-08-05 |
| gpt-oss-20b (free) | OpenAI | $0 | $0 | $0 | $0 | 131.07K | Unranked | 2025-08-05 |
| GLM 4.5 Air (free) | Z.ai | $0 | $0 | $0 | $0 | 131.07K | Unranked | 2025-07-25 |
| Qwen3 Coder 480B A35B (free) | Qwen | $0 | $0 | $0 | $0 | 262K | Unranked | 2025-07-23 |
| Uncensored (free) | Venice | $0 | $0 | $0 | $0 | 32.77K | Unranked | 2025-07-09 |
| Llama 3.3 70B Instruct (free) | Meta | $0 | $0 | $0 | $0 | 65.54K | Unranked | 2024-12-06 |
| Llama 3.2 3B Instruct (free) | Meta | $0 | $0 | $0 | $0 | 131.07K | Unranked | 2024-09-25 |
| Hermes 3 405B Instruct (free) | Nous | $0 | $0 | $0 | $0 | 131.07K | Unranked | 2024-08-16 |
| Mistral Nemo | Mistral | $0.02 | $0.03 | $0.04 | $0.04 | 131.07K | Unranked | 2024-07-19 |
| Llama 3.1 8B Instruct | Meta | $0.02 | $0.05 | $0.04 | $0.04 | 16.38K | Unranked | 2024-07-23 |
| Llama 3 8B Instruct | Meta | $0.04 | $0.04 | $0.06 | $0.06 | 8.19K | Unranked | 2024-04-18 |
| Llama 3 8B Lunaris | Sao10K | $0.04 | $0.05 | $0.07 | $0.07 | 8.19K | Unranked | 2024-08-13 |
| Granite 4.0 Micro | IBM | $0.017 | $0.11 | $0.07 | $0.07 | 131K | Unranked | 2025-10-20 |
| Gemma 3 4B | $0.04 | $0.08 | $0.08 | $0.08 | 131.07K | Unranked | 2025-03-13 | |
| LFM2-24B-A2B | LiquidAI | $0.03 | $0.12 | $0.09 | $0.09 | 32.77K | Unranked | 2026-02-25 |
| Mistral Small 3 | Mistral | $0.05 | $0.08 | $0.09 | $0.09 | 32.77K | Unranked | 2025-01-30 |
| Qwen2.5 7B Instruct | Qwen | $0.04 | $0.1 | $0.09 | $0.09 | 32.77K | Unranked | 2024-10-16 |
| MythoMax 13B | gryphe | $0.06 | $0.06 | $0.09 | $0.09 | 4.1K | Unranked | 2023-07-02 |
| Granite 4.1 8B | IBM | $0.05 | $0.1 | $0.1 | $0.1 | 131.07K | Unranked | 2026-04-30 |
| gpt-oss-20b | OpenAI | $0.03 | $0.14 | $0.1 | $0.1 | 131.07K | Unranked | 2025-08-05 |
| Gemma 3 12B | $0.04 | $0.13 | $0.11 | $0.11 | 131.07K | Unranked | 2025-03-13 | |
| Nova Micro 1.0 | Amazon | $0.035 | $0.14 | $0.11 | $0.11 | 128K | Unranked | 2024-12-05 |
| Command R7B (12-2024) | Cohere | $0.0375 | $0.15 | $0.11 | $0.11 | 128K | Unranked | 2024-12-14 |
| Qwen3.5-9B | Qwen | $0.04 | $0.15 | $0.11 | $0.11 | 262.14K | Unranked | 2026-03-10 |
| Trinity Mini | Arcee AI | $0.045 | $0.15 | $0.12 | $0.12 | 131.07K | Unranked | 2025-12-01 |
| Nemotron Nano 9B V2 | NVIDIA | $0.04 | $0.16 | $0.12 | $0.12 | 131.07K | Unranked | 2025-09-05 |
| Gemma 3n 4B | $0.06 | $0.12 | $0.12 | $0.12 | 32.77K | Unranked | 2025-05-20 | |
| Qwen3 235B A22B Instruct 2507 | Qwen | $0.071 | $0.1 | $0.12 | $0.12 | 262.14K | Unranked | 2025-07-21 |
| Llama 3.2 1B Instruct | Meta | $0.027 | $0.2 | $0.13 | $0.13 | 60K | Unranked | 2024-09-25 |
Pricing FAQ
How is the sample workload cost calculated?
The sample workload uses 1 million input tokens plus 500 thousand output tokens, then applies each model's normalized USD price per 1 million tokens.
Why do input and output token prices matter separately?
Many applications are output-token heavy, while retrieval and classification workloads may be input-token heavy. Comparing both prices helps avoid picking a model that is cheap for the wrong workload shape.
Should I verify prices before production use?
Yes. AI Model Matrix normalizes public pricing metadata for comparison, but provider availability, limits, and prices can change. Always verify the final contract or provider dashboard before production use.
Related Guides
Cheapest LLM APIs
Sort models by estimated workload cost and normalized token prices.
Open guideLargest Context Windows
Find models for long documents, retrieval, and codebase context.
Open guideCoding Models
Compare code-oriented models by cost, context, and popularity rank.
Open guideFree Models
Browse zero-price models for prototypes and evaluation.
Open guideRAG Models
Start from large context windows and practical input-cost constraints.
Open guideChatbot Costs
Find budget-sensitive models for output-heavy assistant traffic.
Open guide