---
title: "Cheapest LLM API in October 2026, by Tier"
description: "Which LLM APIs cost the least per million tokens in October 2026, by tier: frontier, mid, small and open-weight, using each provider's own list prices."
canonical: https://vynaris.com/cheapest-llm-api
author: "Dhrumil Barot"
author_bio: "Dhrumil is an AI inference and LLM infrastructure expert. Dhrumil built Vynaris to make model access, routing and pricing simple and transparent for developers."
published: 2026-10-14
updated: 2026-10-14
---

Pricing guide

# Cheapest LLM API in October 2026, by tier

By [Dhrumil Barot](https://vynaris.com/about#founder), Founder. Dhrumil is an AI inference and LLM infrastructure expert. Dhrumil built Vynaris to make model access, routing and pricing simple and transparent for developers. Published 2026-10-14. Updated 2026-10-14.

The cheapest LLM APIs in October 2026 start at $0.14 per million input tokens for open models on Together AI and $0.10 for small closed models like Gemini 2.5 Flash-Lite, GPT-6 Luna and Claude Haiku 5.5. Mid-tier models run about $0.66 to $2.00 in, and flagships $2.00 to $10.00 in. Every price here is a list price from the provider’s own pricing page, opened on October 14, 2026.

## Key takeaways

- Cheapest per token overall: Llama 3 8B Instruct Lite on Together AI at $0.14 in and $0.14 out per 1M tokens.
- Cheapest small closed model: Gemini 2.5 Flash-Lite (Google) at $0.10 in and $0.40 out.
- Cheapest mid-tier model we checked: DeepSeek V4 Pro 0813 (off-peak) (DeepSeek) at $0.66 in and $1.98 out, off-peak.
- Cheapest flagship for a typical input-heavy month: Grok 4.7 (xAI). Gemini 4 Argon’s $2.00 / $10.00 rate ends November 30, 2026.
- List price is not total cost. Retries, long prompts, caching and tokenizers move the real bill, and [our calculator](https://vynaris.com/calculator) lets you test your own volumes.

## What “cheapest” means (and does not)

We compare the price per million input tokens and per million output tokens at standard speed, short context, uncached and not batched. That is the number every provider prints, so it is the fairest common baseline. It is not the whole bill. Cached input is usually far cheaper (GPT-6 Luna’s cached input is $0.01 per million, for example), batch endpoints are about half price on the models where they apply, and long prompts can switch to a higher rate. The price of the cheapest model is also irrelevant if it cannot do the job.

For cost per finished task rather than per token, see our [cost-per-task pricing comparison](https://vynaris.com/blog/llm-api-pricing-august-2026-cost-per-task) and the [budget-tier floor comparison](https://vynaris.com/blog/gpt-56-luna-vs-haiku-45-vs-flash-lite-vs-deepseek-budget-tier-floor). Our own [pricing table](https://vynaris.com/models) sorts 30 models from cheapest to most expensive, as of July 2026, and the [pricing page](https://vynaris.com/pricing) explains what Vynaris adds on top.

## How we compared

- **Source:** the provider’s own pricing page, linked in every row. No aggregator numbers.
- **Unit:** USD per 1M input tokens and per 1M output tokens, standard processing.
- **Tier:** flagship, mid, small and open-weight, using each provider’s own positioning. A model sits in one tier even where labs overlap.
- **Example month:** 10M input tokens plus 2M output tokens, a simple input-heavy workload, so rows can be compared in dollars. Your ratio will differ.
- **Dates:** promotional and time-limited rates are flagged with their end dates.

As of October 2026. We did not test answer quality, so a lower price here does not mean a model is a good substitute for a higher-priced one.

## 1. Frontier: flagship models from each lab

Each lab's top model by its own pricing page. These are the most expensive per token and the ones most agent traffic does not need on every call.

Cheapest in this tier (example month): Grok 4.7 from xAI, $2.00 in and $6.00 out per 1M tokens.

| Model | Provider | Input / 1M | Output / 1M | Example month | Notes |
| --- | --- | --- | --- | --- | --- |
| Grok 4.7 | xAI | $2.00 | $6.00 | $32.00 | Prompts under 200k tokens. Prompts of 200k or more are $4.00 / $12.00. [source](https://docs.x.ai/developers/pricing) |
| Gemini 4 Argon | Google | $2.00 | $10.00 | $40.00 | Introductory rate through Nov 30, 2026. Google lists $4.00 / $20.00 from Dec 1, 2026. [source](https://ai.google.dev/gemini-api/docs/pricing) |
| Claude Opus 5.5 | Anthropic | $4.00 | $20.00 | $80.00 | [source](https://platform.claude.com/docs/en/about-claude/pricing) |
| Claude Fable 5.1 | Anthropic | $10.00 | $50.00 | $200.00 | [source](https://platform.claude.com/docs/en/about-claude/pricing) |
| GPT-6 Astra | OpenAI | $10.00 | $50.00 | $200.00 | [source](https://developers.openai.com/api/docs/pricing) |

Best suited for: Hard reasoning, long-horizon agent work and code review where one wrong answer costs more than the token bill.

## 2. Mid: balanced models for everyday production work

The workhorse tier: strong enough for most coding, writing and tool use at a fraction of flagship prices.

Cheapest in this tier (example month): DeepSeek V4 Pro 0813 (off-peak) from DeepSeek, $0.66 in and $1.98 out per 1M tokens.

| Model | Provider | Input / 1M | Output / 1M | Example month | Notes |
| --- | --- | --- | --- | --- | --- |
| DeepSeek V4 Pro 0813 (off-peak) | DeepSeek | $0.66 | $1.98 | $10.56 | Off-peak rate. Peak hours (01:00 to 04:00 and 06:00 to 10:00 UTC, Monday to Friday) are $1.32 / $3.96. [source](https://api-docs.deepseek.com/quick_start/pricing) |
| Gemini 3.8 Flash | Google | $0.75 | $3.75 | $15.00 | Rate through Dec 31, 2026. Google lists $1.50 / $7.50 from Jan 1, 2027. [source](https://ai.google.dev/gemini-api/docs/pricing) |
| Grok 4.3 | xAI | $1.25 | $2.50 | $17.50 | Prompts under 200k tokens. [source](https://docs.x.ai/developers/pricing) |
| Claude Sonnet 5.5 | Anthropic | $2.00 | $10.00 | $40.00 | [source](https://platform.claude.com/docs/en/about-claude/pricing) |
| GPT-6.1 Sol | OpenAI | $2.00 | $10.00 | $40.00 | Prompts above the short-context limit are $4.00 / $15.00. [source](https://developers.openai.com/api/docs/pricing) |

Best suited for: General production assistants, coding help and RAG answers where flagship quality is a luxury.

## 3. Small: low-latency models for high volume

The cheapest closed models from the big labs, built for classification, extraction, routing and short answers.

Cheapest in this tier (example month): Gemini 2.5 Flash-Lite from Google, $0.10 in and $0.40 out per 1M tokens.

| Model | Provider | Input / 1M | Output / 1M | Example month | Notes |
| --- | --- | --- | --- | --- | --- |
| Gemini 2.5 Flash-Lite | Google | $0.10 | $0.40 | $1.80 | Text, image and video input. [source](https://ai.google.dev/gemini-api/docs/pricing) |
| GPT-6 Luna | OpenAI | $0.10 | $0.50 | $2.00 | [source](https://developers.openai.com/api/docs/pricing) |
| Claude Haiku 5.5 | Anthropic | $0.10 | $0.50 | $2.00 | Prompts up to 100,000 tokens. Longer prompts are $0.50 / $2.50. [source](https://platform.claude.com/docs/en/about-claude/pricing) |
| DeepSeek V4.1 Flash (off-peak) | DeepSeek | $0.15 | $0.60 | $2.70 | Off-peak rate. Peak hours are $0.30 / $1.20. [source](https://api-docs.deepseek.com/quick_start/pricing) |
| Gemini 3.1 Flash-Lite | Google | $0.25 | $1.50 | $5.50 | Text, image and video input. [source](https://ai.google.dev/gemini-api/docs/pricing) |

Best suited for: Classification, extraction, tagging, routing and any step a short prompt can solve.

## 4. Open-weight: hosted open models

Open models served by Together AI at serverless per-token prices. Other hosts price the same models differently, so check a few before you commit.

Cheapest in this tier (example month): Llama 3 8B Instruct Lite from Together AI, $0.14 in and $0.14 out per 1M tokens.

| Model | Provider | Input / 1M | Output / 1M | Example month | Notes |
| --- | --- | --- | --- | --- | --- |
| Llama 3 8B Instruct Lite | Together AI | $0.14 | $0.14 | $1.68 | [source](https://www.together.ai/pricing) |
| DeepSeek V4 Flash 0731 | Together AI | $0.14 | $0.28 | $1.96 | [source](https://www.together.ai/pricing) |
| GLM-5.3-Flash | Together AI | $0.15 | $0.50 | $2.50 | [source](https://www.together.ai/pricing) |
| gpt-oss-120B | Together AI | $0.15 | $0.60 | $2.70 | [source](https://www.together.ai/pricing) |
| Gemma 4 31B | Together AI | $0.39 | $0.97 | $5.84 | [source](https://www.together.ai/pricing) |

Best suited for: High-volume jobs that run fine on open models and where price per token is the main constraint.

## All tiers side by side

As of October 2026. Example month: 10M input plus 2M output tokens at list price.

| Tier | Cheapest model | Input / 1M | Output / 1M | Example month |
| --- | --- | --- | --- | --- |
| Frontier | Grok 4.7 (xAI) | $2.00 | $6.00 | $32.00 |
| Mid | DeepSeek V4 Pro 0813 (off-peak) (DeepSeek) | $0.66 | $1.98 | $10.56 |
| Small | Gemini 2.5 Flash-Lite (Google) | $0.10 | $0.40 | $1.80 |
| Open-weight | Llama 3 8B Instruct Lite (Together AI) | $0.14 | $0.14 | $1.68 |

The table covers 20 models across four tiers. It is a sample of widely used models, not a complete index of every API.

## Which one fits your use case

| If your job is | Start with |
| --- | --- |
| Classification, tagging or extraction at high volume | The small tier, or an open model at $0.14 in |
| A chat or coding assistant for real users | The mid tier, with a flagship as fallback |
| Hard multi-step agent work | The frontier tier, with cheaper models for the easy steps |
| Batch jobs that can wait | Any tier, on the batch or off-peak endpoint |
| A bill made of mixed tasks you do not want to sort by hand | A router such as Vynaris, which starts at frontier quality and routes down only when evals clear a cheaper model |

Comparing gateways instead of models? See the [best OpenRouter alternatives](https://vynaris.com/compare/openrouter-alternatives), and read [when an LLM router is not worth it](https://vynaris.com/blog/when-not-to-use-llm-router-cost-break-even).

## FAQ

### What is the cheapest LLM API per million tokens?

Among the models we checked on October 14, 2026, the lowest list prices were on Together AI: Llama 3 8B Instruct Lite at $0.14 in and $0.14 out, and DeepSeek V4 Flash 0731 at $0.14 in and $0.28 out. Among the big labs, Gemini 2.5 Flash-Lite is the lowest at $0.10 in and $0.40 out.

### What is the cheapest frontier model API?

Of the flagships we compared, Gemini 4 Argon is $2.00 in and $10.00 out through November 30, 2026, and Grok 4.7 is $2.00 in and $6.00 out. Grok 4.7 is already cheaper on output, and it is cheaper on both from December 1, when Google lists Argon at $4.00 and $20.00.

### Is the cheapest model per token the cheapest per task?

Not always. A cheap model that needs retries, longer prompts or a human fix can cost more per finished task than a pricier one. Anthropic also notes its Claude 4.7 and later models use a tokenizer that produces about 30% more tokens for the same text, so compare tokens as well as rates.

### How can I pay less than list price?

Use batch endpoints, cached input and off-peak windows where the provider offers them. OpenAI and Google list batch prices at half the standard rate on the models above, and DeepSeek charges half price off-peak. A gateway that routes simple requests to a cheaper model can cut the bill further.

### Does Vynaris add a markup on these prices?

Vynaris bills the provider list price for the model that served the request plus a 3% fee on your first $500 of monthly usage and 1% after that. On $1,000 of monthly model usage that is a $20 fee. Each response carries a receipt with the served model and exact cost.

### How often do LLM prices change?

Often, and some of the prices above are promotional with a stated end date. Google's Gemini 4 Argon and Gemini 3.8 Flash rates both rise on dates in the next three months. Check the linked provider page before you budget, and use the Vynaris calculator to model your own volumes.

## Pay list price, with a receipt

Vynaris bills the provider’s list price for the model that served each request, plus a 3% fee on your first $500 a month and 1% after. [Get your API key](https://app.vynaris.com/signin) and see the exact cost of every call, or model your volumes first in the [calculator](https://vynaris.com/calculator).