# Cheapest LLMs by context window size

As of 2026-10-12: the cheapest LLM with at least a 32k-token context window is groq/llama-3.1-8b-instant at $0.0575 per million tokens (blended 3:1 input:output: $0.05 input, $0.08 output). 368 models tracked. Prices are USD per million tokens; free previews and deprecated models are left out.

## At least 32k tokens of context

| Model | Blended | Input | Output | Context tokens |
|---|---|---|---|---|
| groq/llama-3.1-8b-instant | $0.0575 | $0.05 | $0.08 | 131072 |
| cohere/command-r7b-12-2024 | $0.065625 | $0.0375 | $0.15 | 128000 |
| cohere/command-r7b-arabic-02-2025 | $0.065625 | $0.0375 | $0.15 | 128000 |
| cerebras/llama3.1-8b | $0.1 | $0.1 | $0.1 | 128000 |
| mistral/ministral-3-3b-2512 | $0.1 | $0.1 | $0.1 | 131072 |

249 models match; the API returns the top 10 with tool, vision and deprecation fields.

## At least 128k tokens of context

| Model | Blended | Input | Output | Context tokens |
|---|---|---|---|---|
| groq/llama-3.1-8b-instant | $0.0575 | $0.05 | $0.08 | 131072 |
| cohere/command-r7b-12-2024 | $0.065625 | $0.0375 | $0.15 | 128000 |
| cohere/command-r7b-arabic-02-2025 | $0.065625 | $0.0375 | $0.15 | 128000 |
| cerebras/llama3.1-8b | $0.1 | $0.1 | $0.1 | 128000 |
| mistral/ministral-3-3b-2512 | $0.1 | $0.1 | $0.1 | 131072 |

237 models match; the API returns the top 10 with tool, vision and deprecation fields.

## At least 200k tokens of context

| Model | Blended | Input | Output | Context tokens |
|---|---|---|---|---|
| openai/gpt-5-nano | $0.1375 | $0.05 | $0.4 | 400000 |
| openai/gpt-5-nano-2025-08-07 | $0.1375 | $0.05 | $0.4 | 272000 |
| mistral/devstral-small-latest | $0.15 | $0.1 | $0.3 | 256000 |
| mistral/ministral-3-8b-2512 | $0.15 | $0.15 | $0.15 | 262144 |
| mistral/ministral-8b-2512 | $0.15 | $0.15 | $0.15 | 262144 |

178 models match; the API returns the top 10 with tool, vision and deprecation fields.

## At least 1M tokens of context

| Model | Blended | Input | Output | Context tokens |
|---|---|---|---|---|
| google/gemini-2.5-flash-lite | $0.175 | $0.1 | $0.4 | 1048576 |
| openai/gpt-4.1-nano-2025-04-14 | $0.175 | $0.1 | $0.4 | 1047576 |
| anthropic/claude-haiku-5-5 | $0.2 | $0.1 | $0.5 | 1000000 |
| openai/gpt-6-luna | $0.2 | $0.1 | $0.5 | 1050000 |
| openai/gpt-5.6-luna | $0.45 | $0.2 | $1.2 | 1050000 |

96 models match; the API returns the top 10 with tool, vision and deprecation fields.

## Your own constraints from the API

- `POST /v1/models/cheapest` with `{"min_context":128000,"tools":true,"max_usd_per_mtok":2}`: the cheapest models that fit, with how many each rule left out, signed, $0.008 per call.
- `POST /v1/models/estimate`: the cost of a request for given token counts, $0.002.
- `GET /v1/models/{provider}/{model}`: one model's prices, context window and deprecation date, $0.002.
- Paid with x402 in USDC; no account, no API key. Free: [/v1/models/preview](https://g5joilnmeep4ntz3nwezxt5dmu.srv.us/v1/models/preview) (a signed sample).

## Source

Pricing data derived from LiteLLM, Copyright (c) 2023 Berri AI, MIT License. Community-maintained; may lag vendor pages. Data from models.dev (https://models.dev), Copyright (c) 2025 models.dev, MIT License.

Snapshot `models-20261010T070852Z-7d710157`, data as of 2026-10-10T07:08:52Z, source last checked 2026-10-11T19:48:26.000Z.

Licences and attribution: [/licences](https://g5joilnmeep4ntz3nwezxt5dmu.srv.us/licences). How snapshots are made and verified: [/provenance](https://g5joilnmeep4ntz3nwezxt5dmu.srv.us/provenance).

Data is provided as is, without warranty of any kind. Payments are final. Every dataset is normalised (modified) from the sources named here, and no source endorses this service.