The LLM Price Index
Every major large language model, normalized to dollars per million tokens. Scored on value, cheapness, and frontier capability. Independent, automated, refreshed daily.
What this is
The LLM Price Index is a live, independently maintained price comparison for every large language model offered through the OpenRouter catalog. Input and output pricing is normalized to dollars per million tokens, blended 3:1 to produce a single comparable figure, and scored 0–10 on three axes: value, cheapness, and frontier capability. The cheapest paid model right now is Gemma 3n 4B at $0.025 per million tokens (3:1 blended).
Qwen3 Next 80B A3B Instruct (free)
7.9/10Qwen3-Next-80B-A3B-Instruct is an instruction-tuned chat model in the Qwen3-Next series optimized for fast, stable responses without “thinking” traces. It targets complex tasks…
LFM2.5-1.2B-Thinking (free)
7.8/10LFM2.5-1.2B-Thinking is a lightweight reasoning-focused model optimized for agentic tasks, data extraction, and RAG—while still running comfortably on edge devices. It supports…
Nemotron Nano 9B V2 (free)
7.8/10NVIDIA-Nemotron-Nano-9B-v2 is a large language model (LLM) trained from scratch by NVIDIA, and designed as a unified model for both reasoning and…
gpt-oss-120b (free)
7.8/10gpt-oss-120b is an open-weight, 117B-parameter Mixture-of-Experts (MoE) language model from OpenAI designed for high-reasoning, agentic, and general-purpose production use cases. It activates…
Trinity Mini
7.5/10Trinity Mini is a 26B-parameter (3B active) sparse mixture-of-experts language model featuring 128 experts with 8 active per token. Engineered for efficient…
Llama 3.2 3B Instruct (free)
7.5/10Llama 3.2 3B is a 3-billion-parameter multilingual large language model, optimized for advanced natural language processing tasks like dialogue generation, reasoning, and…
Hermes 3 405B Instruct (free)
7.5/10Hermes 3 is a generalist language model with many improvements over Hermes 2, including advanced agentic capabilities, much better roleplaying, reasoning, multi-turn…
Trinity Large Thinking
7.2/10Trinity Large Thinking is a powerful open source reasoning model from the team at Arcee AI. It shows strong performance in PinchBench,…
Nemotron Nano 9B V2
7.2/10NVIDIA-Nemotron-Nano-9B-v2 is a large language model (LLM) trained from scratch by NVIDIA, and designed as a unified model for both reasoning and…
gpt-oss-120b
7.1/10gpt-oss-120b is an open-weight, 117B-parameter Mixture-of-Experts (MoE) language model from OpenAI designed for high-reasoning, agentic, and general-purpose production use cases. It activates…
Qwen Plus 0728 (thinking)
7.1/10Qwen Plus 0728, based on the Qwen3 foundation model, is a 1 million context hybrid reasoning model with a balanced performance, speed,…
Qwen Plus 0728
7.1/10Qwen Plus 0728, based on the Qwen3 foundation model, is a 1 million context hybrid reasoning model with a balanced performance, speed,…
Qwen3 30B A3B Instruct 2507
7.1/10Qwen3-30B-A3B-Instruct-2507 is a 30.5B-parameter mixture-of-experts language model from Qwen, with 3.3B active parameters per inference. It operates in non-thinking mode and is…
gpt-oss-safeguard-20b
6.7/10gpt-oss-safeguard-20b is a safety reasoning model from OpenAI built upon gpt-oss-20b. This open-weight, 21B-parameter Mixture-of-Experts (MoE) model offers lower latency for safety…
Command R7B (12-2024)
6.6/10Command R7B (12-2024) is a small, fast update of the Command R+ model, delivered in December 2024. It excels at RAG, tool…
Qwen3 30B A3B Thinking 2507
6.6/10Qwen3-30B-A3B-Thinking-2507 is a 30B parameter Mixture-of-Experts reasoning model optimized for complex tasks requiring extended multi-step thinking. The model is designed specifically for…
Mercury 2
6.3/10Mercury 2 is an extremely fast reasoning LLM, and the first reasoning diffusion LLM (dLLM). Instead of generating tokens sequentially, Mercury 2…
Qwen3 14B
6.3/10Qwen3-14B is a dense 14.8B parameter causal language model from the Qwen3 series, designed for both complex reasoning and efficient dialogue. It…
Qwen3 Next 80B A3B Instruct
6.2/10Qwen3-Next-80B-A3B-Instruct is an instruction-tuned chat model in the Qwen3-Next series optimized for fast, stable responses without “thinking” traces. It targets complex tasks…
MiniMax M1
6.1/10MiniMax-M1 is a large-scale, open-weight reasoning model designed for extended context and high-efficiency inference. It leverages a hybrid Mixture-of-Experts (MoE) architecture paired…
DeepSeek V3.2
6.1/10DeepSeek-V3.2 is a large language model designed to harmonize high computational efficiency with strong reasoning and agentic tool-use performance. It introduces DeepSeek…
Qwen3 32B
6.1/10Qwen3-32B is a dense 32.8B parameter causal language model from the Qwen3 series, optimized for both complex reasoning and efficient dialogue. It…
Rnj 1 Instruct
6.1/10Rnj-1 is an 8B-parameter, dense, open-weight model family developed by Essential AI and trained from scratch with a focus on programming, math,…
Llama 3.3 Nemotron Super 49B V1.5
6.0/10Llama-3.3-Nemotron-Super-49B-v1.5 is a 49B-parameter, English-centric reasoning/chat model derived from Meta’s Llama-3.3-70B-Instruct with a 128K context. It’s post-trained for agentic workflows (RAG, tool…
Qwen3 30B A3B
5.9/10Qwen3, the latest generation in the Qwen large language model series, features both dense and mixture-of-experts (MoE) architectures to excel in reasoning,…
Qwen3 Next 80B A3B Thinking
5.9/10Qwen3-Next-80B-A3B-Thinking is a reasoning-first chat model in the Qwen3-Next line that outputs structured “thinking” traces by default. It’s designed for hard multi-step…
Qwen3 8B
5.9/10Qwen3-8B is a dense 8.2B parameter causal language model from the Qwen3 series, designed for both reasoning-heavy tasks and efficient dialogue. It…
GLM 4.7
5.8/10GLM-4.7 is Z.ai’s latest flagship model, featuring upgrades in two key areas: enhanced programming capabilities and more stable multi-step reasoning/execution. It demonstrates…
Kimi K2 Thinking
5.8/10Kimi K2 Thinking is Moonshot AI’s most advanced open reasoning model to date, extending the K2 series into agentic, long-horizon reasoning. Built…
MiniMax M2
5.7/10MiniMax-M2 is a compact, high-efficiency large language model optimized for end-to-end coding and agentic workflows. With 10 billion activated parameters (230 billion…
Command R (08-2024)
5.6/10command-r-08-2024 is an update of the [Command R](/models/cohere/command-r) with improved performance for multilingual retrieval-augmented generation (RAG) and tool use. More broadly, it…
Grok 3 Mini
5.6/10A lightweight model that thinks before responding. Fast, smart, and great for logic-based tasks that do not require deep domain knowledge. The…
Olmo 2 32B Instruct
5.5/10OLMo-2 32B Instruct is a supervised instruction-finetuned variant of the OLMo-2 32B March 2025 base model. It excels in complex reasoning and…
Qwen3 Max Thinking
5.5/10Qwen3-Max-Thinking is the flagship reasoning model in the Qwen3 series, designed for high-stakes cognitive tasks that require deep, multi-step reasoning. By significantly…
ERNIE 4.5 21B A3B Thinking
5.4/10ERNIE-4.5-21B-A3B-Thinking is Baidu's upgraded lightweight MoE model, refined to boost reasoning depth and quality for top-tier performance in logical puzzles, math, science,…
Qwen3 235B A22B Thinking 2507
5.3/10Qwen3-235B-A22B-Thinking-2507 is a high-performance, open-weight Mixture-of-Experts (MoE) language model optimized for complex reasoning tasks. It activates 22B of its 235B parameters per…
DeepSeek R1T2 Chimera
5.3/10DeepSeek-TNG-R1T2-Chimera is the second-generation Chimera model from TNG Tech. It is a 671 B-parameter mixture-of-experts text-generation model assembled from DeepSeek-AI’s R1-0528, R1,…
QwQ 32B
5.2/10QwQ is the reasoning model of the Qwen series. Compared with conventional instruction-tuned models, QwQ, which is capable of thinking and reasoning,…
Hunyuan A13B Instruct
5.2/10Hunyuan-A13B is a 13B active parameter Mixture-of-Experts (MoE) language model developed by Tencent, with a total parameter count of 80B and support…
Llama 3 8B Lunaris
5.1/10Lunaris 8B is a versatile generalist and roleplaying model based on Llama 3. It's a strategic merge of multiple models, designed to…
DeepSeek V3.1
5.0/10DeepSeek-V3.1 is a large hybrid reasoning model (671B parameters, 37B active) that supports both thinking and non-thinking modes via prompt templates. It…
Hermes 4 70B
5.0/10Hermes 4 70B is a hybrid reasoning model from Nous Research, built on Meta-Llama-3.1-70B. It introduces the same hybrid mode as the…
Grok 3 Mini Beta
5.0/10Grok 3 Mini is a lightweight, smaller thinking model. Unlike traditional models that generate answers immediately, Grok 3 Mini thinks before responding.…
Qwen3 Max
5.0/10Qwen3-Max is an updated release built on the Qwen3 series, offering major improvements in reasoning, instruction following, multilingual support, and long-tail knowledge…
Devstral Medium
4.9/10Devstral Medium is a high-performance code generation and agentic reasoning model developed jointly by Mistral AI and All Hands AI. Positioned as…
R1 0528
4.8/10May 28th update to the [original DeepSeek R1](/deepseek/deepseek-r1) Performance on par with [OpenAI o1](/openai/o1), but open-sourced and with fully open reasoning tokens.…
DeepSeek V3.2 Speciale
4.8/10DeepSeek-V3.2-Speciale is a high-compute variant of DeepSeek-V3.2 optimized for maximum reasoning and agentic performance. It builds on DeepSeek Sparse Attention (DSA) for…
o3 Mini High
4.8/10OpenAI o3-mini-high is the same model as [o3-mini](/openai/o3-mini) with reasoning_effort set to high. o3-mini is a cost-efficient language model optimized for STEM…
Free · No spam
Price drops, new models, deprecations
Every Tuesday: which models got cheaper, which launched, which got pulled. Five minutes. No filler.
Three axes, one overall score
- Value (35%) — capability per dollar. Context length, vision, tools, and structured-output support divided by log-scaled blended price.
- Cheapness (35%) — raw affordability. Free models score 10. Paid models use an inverse log curve anchored at $0.01 / Mtok.
- Frontier (30%) — how close to the state of the art. Recent releases, long context windows, and premium pricing all contribute.
Blended price formula
- Most production workloads are input-heavy, so the index uses a 3:1 blended price:
(input × 0.75) + (output × 0.25). - All prices are normalized to dollars per million tokens. OpenRouter publishes per-token figures which we multiply by 1,000,000 before display.
Where does the pricing data come from?
Every model and price on this page is sourced from OpenRouter's public models API, which aggregates pricing directly from model providers including Anthropic, OpenAI, Google, Mistral, Meta, xAI, DeepSeek, and dozens of others. The pipeline re-fetches and re-scores the entire catalog once per day.
Why normalize to $/million tokens?
Model providers publish prices in inconsistent units — per 1K tokens, per million tokens, per character, sometimes per request. Comparing them directly is error-prone. Dollars per million tokens is the industry's most common reporting unit and makes cross-provider comparisons immediate and honest.
What does "3:1 blended" mean?
Most production LLM workloads are input-heavy — context, RAG retrievals, system prompts — while output is comparatively short. A 3:1 input:output ratio is the informal industry convention for producing a single number that reflects typical cost: (input × 0.75) + (output × 0.25). Your actual ratio may differ; always check both input and output columns for workloads with long generations.
What's the cheapest LLM right now?
The cheapest paid model as of the latest scan is Gemma 3n 4B from Google at $0.025 per million tokens (3:1 blended). Sort by "Cheapest" above for the full ranking. Many providers also offer free-tier variants of their models, which score a perfect 10 on the cheapness axis.
Is this affiliated with OpenRouter or any provider?
No. MegaOne AI is independent. OpenRouter is used as a public data source because their models API is the most complete and up-to-date LLM catalog available, but this directory is not operated by OpenRouter and we rate all models — including ones that compete with one another.
How often does the price index update?
A full re-fetch, re-score, and daily snapshot runs once per 24 hours. Snapshots are written to a history table so we can build price-over-time charts and detect drops. New models typically appear within 24 hours of being added to OpenRouter.
Is it free to use?
Yes. Browsing, filtering, sorting, and searching the entire price index is free. The weekly email briefing is free. There is no account required and no paywall.