State of AI Models 2025 — Data-Driven Report

4,587

Total Models

Providers

Free Models

527

Open-Weight

2,350

Tool Calling

1,306

Reasoning

1,487

Vision

2,195

128K+ Context

1. Provider Landscape

The AI model ecosystem spans 95 providers, from tech giants to specialized startups. The top 15 providers account for the majority of models:

Provider	Models	Notable Models
OpenRouter	415	Aggregator — routes to 100+ models
Google	261	Gemini 2.5 Pro/Flash, Gemma 3
Requesty	234	Aggregator — unified API
Cohere	197	Command R+, Embed v3
xAI	193	Grok 3, Grok 3 Mini
DeepSeek	184	DeepSeek R1, V3
Meta	163	Llama 4 Maverick/Scout
Mistral	155	Mistral Large, Codestral
Alibaba (Qwen)	139	Qwen3-235B, QwQ
Anthropic	121	Claude Sonnet 4, Opus 4
OpenAI	115	GPT-4.1, o3, o4-mini
Microsoft	99	Phi-4, Florence 2
Amazon	96	Nova Pro, Titan
NVIDIA	87	Nemotron, Llama Nemotron
01.ai	83	Yi-Lightning, Yi-VL

Key Insight: Aggregators (OpenRouter, Requesty) offer the widest selection but may duplicate models available from first-party providers. For the best pricing, go direct to the source.

2. Pricing Distribution

AI model pricing varies dramatically — from completely free to over $15 per million input tokens. Here is the breakdown of the 4,587 models:

Free

81 models

< $0.50/M

~1,800 models

$0.50–5/M

~1,400 models

> $5/M

~480 models

Key Insight: The median input price for tool-calling models is $0.50/M tokens, while reasoning models median is $0.80/M. Vision-capable models average $1.50/M — still affordable for most production use cases.

3. Capability Adoption

Modern AI models increasingly support advanced capabilities beyond basic text generation:

Capability	Models	% of Total	Avg Input $/M
Tool Calling	2,350	51.2%	$1.50
Reasoning	1,306	28.5%	$2.10
Structured Output	829	18.1%	$1.80
Vision (Image Input)	1,487	32.4%	$1.50
Open Weights	527	11.5%	Free or low-cost
Image Generation	28	0.6%	$3.00+
Audio Input	118	2.6%	$2.50+
Audio Output	34	0.7%	$3.00+
Video Input	167	3.6%	$2.00+

Key Insight: Over half of all models now support tool calling — it has become table stakes for production AI. Reasoning capabilities are growing fast, with 1,306 models (28.5%) supporting extended thinking.

4. Context Window Revolution

Context windows have grown exponentially. The average context window across all models is now approximately 200K tokens:

< 32K

~800 models

32K–128K

~1,000 models

128K–1M

~2,195 models

1M+

~30 models

Largest Context Windows

Model	Context	Provider
Google Gemini 2.5 Pro	1,048,576	Google
Google Gemini 2.5 Flash	1,048,576	Google
Meta Llama 4 Scout	10,000,000	Meta
Meta Llama 4 Maverick	1,048,576	Meta
Google Gemma 3 27B	131,072	Google

Key Insight: 128K+ context is now the norm — 2,195 models (47.8%) support it. Meta's Llama 4 Scout leads with a 10M token window, making entire codebases and books processable in a single prompt.

5. The Rise of Free & Open-Source AI

81 models are completely free to use, and 527 have open weights. Here are the most capable free models:

Model	Context	Capabilities	Provider
Google Gemini 2.5 Flash	1M	TC, Reasoning, Vision, SO	Google
DeepSeek R1	128K	Reasoning, TC	DeepSeek
Meta Llama 4 Maverick	1M	TC, Vision	Meta
Alibaba Qwen3-235B	128K	TC, Reasoning, SO	Alibaba
Google Gemma 3 27B	131K	Vision, TC	Google

Key Insight: Free models now rival paid ones in capability. Google Gemini 2.5 Flash (free tier) offers 1M context, tool calling, reasoning, and vision — making it viable for production use at zero cost.

6. Best Value Models by Use Case

Use Case	Best Free	Best Paid (Cheapest)	Best Overall
General Chat	Gemini 2.5 Flash	DeepSeek V3 ($0.07/$0.28)	Claude Sonnet 4
Coding	DeepSeek R1	DeepSeek V3 ($0.07/$0.28)	Claude Sonnet 4
AI Agents	Gemini 2.5 Flash	Grok 3 Mini ($0.30/$0.50)	Claude Sonnet 4
Reasoning	DeepSeek R1	Grok 3 Mini ($0.30/$0.50)	o3
Vision	Gemini 2.5 Flash	Gemma 3 4B (free)	Gemini 2.5 Pro
Large Context	Llama 4 Scout (10M)	Gemini 2.5 Flash ($0.15/$0.60)	Gemini 2.5 Pro

7. Key Trends & Predictions

Trend 1: Agentic AI is the new default. 51% of models support tool calling, and 1,080 models are classified as "agentic" (tool_call + chat). Expect this to reach 80%+ by 2026.

Trend 2: Context windows are commoditized. 128K context is now standard. 1M+ context models are growing, with Google and Meta leading. Expect 10M+ to become common by 2026.

Trend 3: Free tiers are production-ready. 81 free models with capabilities like tool calling and reasoning mean that cost is no longer a barrier to entry for AI development.

Trend 4: Multimodal is mainstream. 1,548 models support more than text input. Vision (1,487 models) is nearly universal among flagship models. Audio and video are the next frontiers.

Trend 5: Open weights are accelerating. 527 open-weight models exist, with Meta's Llama 4 and Alibaba's Qwen3 leading. Expect open-source to match proprietary capabilities within 6 months.

📊 State of AI Models 2025