Cheapest LLMs for agentic workloads
Mistral: Mistral Nemo is the cheapest option at $0.019 per million input tokens. Agents make many calls per task, so per-token price compounds fast. These models support tool calling and structured output, the two things an agent loop cannot run without.
252 models qualify
Ranked by blended cost - input weighted 75%, output 25%.
| # | Model | Blended / 1M | Input / 1M | Output / 1M | Context | Capabilities |
|---|---|---|---|---|---|---|
| 1 | Auto Router (Beta) OpenRouter | $-1000000.0000 | $-1000000.0000 | $-1000000.0000 | 2M | ReasoningToolsVision |
| 2 | Auto Router OpenRouter | $-1000000.0000 | $-1000000.0000 | $-1000000.0000 | 2M | ReasoningToolsVision |
| 3 | Nex-N2.5-Mini (free) Nex AGI | Free | Free | Free | 262K | Free tierReasoningToolsVisionOpen weights |
| 4 | Nex-N2.5-Pro (free) Nex AGI | Free | Free | Free | 262K | Free tierReasoningToolsVisionOpen weights |
| 5 | Dots3-Note Preview (free) Dots Studio | Free | Free | Free | 512K | Free tierReasoningToolsVision |
| 6 | LFM2.5-2.6B (free) Liquid AI | Free | Free | Free | 66K | Free tierReasoningToolsOpen weights |
| 7 | Free Models Router OpenRouter | Free | Free | Free | 200K | Free tierReasoningToolsVision |
| 8 | Mistral Nemo Mistral AI | $0.022 | $0.019 | $0.030 | 131K | ToolsOpen weights |
| 9 | Ling 3.0 Flash inclusionAI | $0.032 | $0.021 | $0.063 | 262K | ReasoningToolsOpen weights |
| 10 | Qwen3.7 Flash Qwen | $0.055 | $0.030 | $0.130 | 1M | ReasoningToolsVision |
| 11 | gpt-oss-20b OpenAI | $0.055 | $0.030 | $0.130 | 131K | ReasoningToolsOpen weights |
| 12 | Llama 3.1 8B Instruct Meta Llama | $0.057 | $0.050 | $0.080 | 131K | ToolsOpen weights |
| 13 | Mercury 2.5 Inception | $0.068 | $0.040 | $0.150 | 260K | ReasoningTools |
| 14 | gpt-oss-120b OpenAI | $0.070 | $0.037 | $0.170 | 131K | ReasoningToolsOpen weights |
| 15 | DeepSeek V4 Flash 0731 DeepSeek | $0.075 | $0.060 | $0.120 | 1.3M | ReasoningToolsOpen weights |
| 16 | Gemma 3 12B | $0.075 | $0.050 | $0.150 | 131K | ToolsVisionOpen weights |
| 17 | $0.084 | $0.048 | $0.193 | 262K | ToolsOpen weights | |
| 18 | Nemotron 3 Nano 30B A3B NVIDIA | $0.087 | $0.050 | $0.200 | 262K | ReasoningToolsOpen weights |
| 19 | Ling 3.0 Flash VL inclusionAI | $0.090 | $0.060 | $0.180 | 131K | Free tierReasoningToolsVisionOpen weights |
| 20 | Ling 3.0 Flash Fin inclusionAI | $0.090 | $0.060 | $0.180 | 262K | Free tierReasoningTools |
| 21 | Reka Edge Reka AI | $0.100 | $0.100 | $0.100 | 16K | ToolsVisionOpen weights |
| 22 | Ministral 3 3B 2512 Mistral AI | $0.100 | $0.100 | $0.100 | 131K | ToolsVisionOpen weights |
| 23 | Mistral Small 3.2 24B Mistral AI | $0.106 | $0.075 | $0.200 | 256K | ToolsVisionOpen weights |
| 24 | DeepSeek V4 Flash 0423 DeepSeek | $0.107 | $0.086 | $0.171 | 1.0M | ReasoningToolsOpen weights |
| 25 | Granite 4.2 8B IBM Granite | $0.107 | $0.060 | $0.250 | 131K | ReasoningToolsOpen weights |
Frequently asked
What is the cheapest LLM for multi-step agents?
Mistral: Mistral Nemo at $0.019 per million input tokens and $0.030 per million output tokens, with a 131K token context window.
How many models qualify for multi-step agents?
252 of the 340 models tracked here meet the criteria for this list.
How much do prices vary within this category?
By roughly 3,100×, from Mistral Nemo at the bottom to GPT-5.4 Pro at the top.