AI Model Comparison 2026
Context window, API pricing, modalities, and capabilities for the major AI models, side by side. Figures are web-verified against each provider's docs (reviewed June 2026). Pick a model for its full spec sheet and prompting tips.
| Model | Provider | Context | Input /1M | Output /1M | Inputs |
|---|---|---|---|---|---|
| GPT-5.5 | OpenAI | 400K | $5.00 | $30.00 | Text, Images |
| GPT-5.4 nano | OpenAI | 400K | $0.20 | $1.25 | Text, Images |
| Claude Opus 4.8 | Anthropic | 1.0M | $5.00 | $25.00 | Text, Images |
| Claude Sonnet 4.6 | Anthropic | 1.0M | $3.00 | $15.00 | Text, Images |
| Claude Haiku 4.5 | Anthropic | 200K | $1.00 | $5.00 | Text, Images |
| Gemini 3.1 Pro | 1.0M | $2.00 | $12.00 | Text, Images, Audio, Video | |
| Gemini 2.5 Pro | 1.0M | $1.25 | $10.00 | Text, Images, Audio, Video | |
| Gemini 2.5 Flash | 1.0M | $0.30 | $2.50 | Text, Images, Audio, Video | |
| Grok 4.3 | xAI | 1.0M | $1.25 | $2.50 | Text, Images |
| DeepSeek V4 | DeepSeek | 1.0M | $0.43 | $0.87 | Text |
| DeepSeek V3.2 | DeepSeek | 164K | $0.28 | $0.42 | Text |
| Llama 4 Maverick (hosted) | Meta | 1.0M | $0.20 | $0.60 | Text, Images |
| Mistral Large 3 | Mistral | 128K | $0.50 | $1.50 | Text, Images |
Pricing is USD per 1M tokens, web-verified June 2026. Confirm before budgeting: OpenAI · Anthropic · Google · xAI · DeepSeek · Meta · Mistral. Need the cost for your own prompt? Use the token counter & cost calculator.
Quick takeaways
- Largest context: Claude Opus 4.8, Claude Sonnet 4.6, Gemini 3.1 Pro, Gemini 2.5 Pro, Gemini 2.5 Flash, Grok 4.3, DeepSeek V4, Llama 4 Maverick (hosted) — 1.0M tokens.
- Cheapest input: GPT-5.4 nano, Llama 4 Maverick (hosted) at $0.20/1M.
- Open-weight (self-hostable): DeepSeek V4, DeepSeek V3.2, Llama 4 Maverick (hosted), Mistral Large 3.
- Native real-time web: Grok 4.3.
- Multimodal (image input): 11 of 13 models — only DeepSeek V4, DeepSeek V3.2 is text-only.
Frequently asked questions
Which AI model has the largest context window?
Claude Opus 4.8, Claude Sonnet 4.6, Gemini 3.1 Pro, Gemini 2.5 Pro, Gemini 2.5 Flash, Grok 4.3, DeepSeek V4, Llama 4 Maverick (hosted) lead with a 1,000,000-token (1.0M) context window — enough to hold large codebases or several long documents at once.
Which AI model API is cheapest?
On input tokens, GPT-5.4 nano, Llama 4 Maverick (hosted) is the cheapest at $0.20 per 1M tokens. Total cost depends on output length too — output usually costs several times more than input.
Which AI models are open-weight?
DeepSeek V4, DeepSeek V3.2, Llama 4 Maverick (hosted), Mistral Large 3 are open-weight, meaning you can self-host them. The rest (OpenAI, Anthropic, Google, xAI) are API-only.
Which AI model has real-time web access?
Grok 4.3 has native live web/search access, making it the strongest pick for current events and time-sensitive research. Others rely on their training data unless you add a search tool.
How current is this pricing and these specs?
These figures were web-verified against each provider's documentation in June 2026. LLM pricing and model lineups change often, so each model links to its provider source — confirm before budgeting.