The Cheapest AI Models in 2026, Ranked by Price
The cheapest capable model changes every few weeks. These tables are generated from our verified price data, so they are always current.
Every table below ranks current models by blended price: three input tokens for every output token, which suits chat and most business applications. Prices are list prices per 1M tokens from each provider's own pricing page, checked on 13 Oct 2026. If your workload is mostly input (agents, document search) or mostly output (writing, code generation), check the input or output column rather than the blend.
Cheapest from the model makers themselves
These models are sold directly by the lab that made them, which usually means the most reliable service, published rate limits and prompt caching or batch discounts where offered.
| Model | Input | Output | Blended 3:1 | Context |
|---|---|---|---|---|
| Ministral 3 3B Mistral AI | $0.10 | $0.10 | $0.10 | — |
| Ministral 3 8B Mistral AI | $0.15 | $0.15 | $0.15 | — |
| Claude Haiku 5.5 Anthropic | $0.10 | $0.50 | $0.20 | 1M |
| GPT-6 Luna OpenAI | $0.10 | $0.50 | $0.20 | 1.05M |
| Ministral 3 14B Mistral AI | $0.20 | $0.20 | $0.20 | — |
| Qwen3.8 Flash Alibaba (Qwen) | $0.15 | $0.47 | $0.23 | 1M |
| GLM-5.3 Flash Z.ai | $0.15 | $0.50 | $0.24 | 1M |
| Mistral Small 4 Mistral AI | $0.15 | $0.60 | $0.26 | 256K |
| DeepSeek V4.1 Flash DeepSeek | $0.30 | $1.20 | $0.52 | 1M |
| MiniMax M3 MiniMax | $0.30 | $1.20 | $0.52 | 1M |
Current models ranked by blended price per 1M tokens (3 input : 1 output). Verified 13 Oct 2026; updates automatically.
The very cheapest are small models meant for simple, high-volume work: classification, routing, extraction and short answers. For general chat and instruction-following, models such as GPT-6 Luna ($0.10 / $0.50) and DeepSeek V4.1 Flash ($0.30 / $1.20, half that off-peak) are the usual low-cost starting points.
Small and fast models
The cheapest tier, including open-weight models served by third-party hosts. Good for high-volume pipelines and as the "easy request" side of a model router.
| Model | Input | Output | Blended 3:1 | Context |
|---|---|---|---|---|
| gpt-oss-20b OpenAI · via OpenRouter | $0.018 | $0.090 | $0.036 | 128K |
| Nemotron 3.5 Lightning NVIDIA · via OpenRouter | $0.025 | $0.12 | $0.049 | 1M |
| Gemma 4 26B-A4B Google · via OpenRouter | $0.042 | $0.22 | $0.087 | 256K |
| Ministral 3 3B Mistral AI | $0.10 | $0.10 | $0.10 | — |
| Ministral 3 8B Mistral AI | $0.15 | $0.15 | $0.15 | — |
| Gemma 4 31B Google · via OpenRouter | $0.090 | $0.34 | $0.15 | 256K |
| Claude Haiku 5.5 Anthropic | $0.10 | $0.50 | $0.20 | 1M |
| GPT-6 Luna OpenAI | $0.10 | $0.50 | $0.20 | 1.05M |
| Ministral 3 14B Mistral AI | $0.20 | $0.20 | $0.20 | — |
| Qwen3.6 35B-A3B Alibaba (Qwen) · via OpenRouter | $0.050 | $0.70 | $0.21 | 256K |
Current models ranked by blended price per 1M tokens (3 input : 1 output). Verified 13 Oct 2026; updates automatically.
Workhorse models
Mid-priced models for everyday production work: support bots, content, summarisation and most agent steps.
| Model | Input | Output | Blended 3:1 | Context |
|---|---|---|---|---|
| gpt-oss-120b OpenAI · via OpenRouter | $0.030 | $0.15 | $0.060 | 128K |
| Nemotron 3 Super NVIDIA · via OpenRouter | $0.085 | $0.40 | $0.16 | 256K |
| MiniMax M3 MiniMax | $0.30 | $1.20 | $0.52 | 1M |
| Command A+ Cohere | $0.30 | $1.50 | $0.60 | 128K |
| Qwen3.7 Plus Alibaba (Qwen) | $0.40 | $1.60 | $0.70 | 1M |
| Mistral Large 4 Mistral AI | $0.70 | $2.10 | $1.05 | 1M |
| Gemini 3.8 Flash | $0.75 | $3.75 | $1.50 | 1M |
| GPT-6.1 Terra OpenAI | $1.00 | $5.00 | $2.00 | 1.05M |
Current models ranked by blended price per 1M tokens (3 input : 1 output). Verified 13 Oct 2026; updates automatically.
Gemini 3.8 Flash's price is a promotion that runs until 31 December 2026, after which it doubles to $1.50 / $7.50. Claude Sonnet 5.5 and GPT-6.1 Sol, both at $2.00 / $10.00, cost more than everything in this table; the full price table lists the rest of the tier.
Frontier models
Each lab's most capable models. The spread is wide: the most expensive frontier models cost several times as much as the cheapest.
| Model | Input | Output | Blended 3:1 | Context |
|---|---|---|---|---|
| DeepSeek V4 Pro DeepSeek | $1.32 | $3.96 | $1.98 | 1M |
| Muse Spark 1.3 Meta | $1.25 | $4.25 | $2.00 | 1M |
| GLM-5.3 Z.ai | $1.40 | $4.40 | $2.15 | 1M |
| Grok 4.7 xAI | $2.00 | $6.00 | $3.00 | 500K |
| Qwen3.8 Max Alibaba (Qwen) | $2.00 | $6.00 | $3.00 | 1M |
| Gemini 4 Argon | $2.00 | $10.00 | $4.00 | 1M |
| Kimi K3.1 Pro Moonshot AI | $2.00 | $10.00 | $4.00 | 1M |
| Gemini 3.1 Pro | $2.00 | $12.00 | $4.50 | 1M |
| Kimi K3 Moonshot AI | $3.00 | $15.00 | $6.00 | 1M |
| Claude Opus 5.5 Anthropic | $4.00 | $20.00 | $8.00 | 1M |
Current models ranked by blended price per 1M tokens (3 input : 1 output). Verified 13 Oct 2026; updates automatically.
DeepSeek V4 Pro's price is its peak rate; off-peak requests cost half. Grok 4.7's price applies to prompts under 200K tokens. The frontier median is $4.00 per 1M tokens, down 3% since 13 Jun 2026 (see the token price index).
Cheapest open-weight models
Open-weight models can be run on your own hardware or bought from competing hosts. Where a host is named, the price is the cheapest host listed on OpenRouter when we checked; otherwise it is the maker's own price. Speed, reliability and context limits vary from host to host.
| Model | Input | Output | Blended 3:1 | Context |
|---|---|---|---|---|
| gpt-oss-20b OpenAI · via OpenRouter | $0.018 | $0.090 | $0.036 | 128K |
| Nemotron 3.5 Lightning NVIDIA · via OpenRouter | $0.025 | $0.12 | $0.049 | 1M |
| gpt-oss-120b OpenAI · via OpenRouter | $0.030 | $0.15 | $0.060 | 128K |
| Gemma 4 26B-A4B Google · via OpenRouter | $0.042 | $0.22 | $0.087 | 256K |
| Gemma 4 31B Google · via OpenRouter | $0.090 | $0.34 | $0.15 | 256K |
| Nemotron 3 Super NVIDIA · via OpenRouter | $0.085 | $0.40 | $0.16 | 256K |
| Qwen3.6 35B-A3B Alibaba (Qwen) · via OpenRouter | $0.050 | $0.70 | $0.21 | 256K |
| Mistral Small 4 Mistral AI | $0.15 | $0.60 | $0.26 | 256K |
| Qwen3.8 27B Alibaba (Qwen) · via OpenRouter | $0.039 | $1.25 | $0.34 | 256K |
| Muse Glimmer 30B Meta · via OpenRouter | $0.30 | $1.20 | $0.52 | 128K |
Current models ranked by blended price per 1M tokens (3 input : 1 output). Verified 13 Oct 2026; updates automatically.
Running them yourself avoids per-token charges but not costs: the local LLM calculator estimates speed on your hardware and whether it pays back against these hosted prices.
Cheap is only cheap if it works
- Test on your own prompts. A model that needs a second attempt, or a human fix, costs more than a pricier model that gets it right first time.
- Count tokens, not just prices. Tokenizers differ, and reasoning models bill their thinking as output. Two models with the same list price can produce very different bills.
- Look at caching. For agents and chat apps that resend context, the cached-input price can matter more than the headline input price. See how AI pricing works.
- Mix models. Route easy requests to a small model and hard ones to a larger model. More ways to cut costs.
Compare up to four of these models side by side with the comparison tool, or price your exact traffic in the token cost calculator.