ChatGPT, Gemini, Claude, DeepSeek and Grok — full comparison of pricing, context windows and capabilities, covering subscriptions and API pricing.
| Model | Latest | API in | API out | Context | Free tier | Best for |
|---|---|---|---|---|---|---|
ChatGPT OpenAI chatgpt.com ↗ |
GPT-5.6 (Sol/Terra/Luna) | $1–5 |
$6–30 |
272K | Free tier | General / Writing / Coding |
Gemini Google gemini.google.com ↗ |
Gemini 3 Pro / 3.1 | $2 |
$12 |
1M | AI Studio free tier | Multimodal / Search / Google |
Claude Anthropic claude.ai ↗ |
Opus 4.5 / Sonnet 5 | $3–5 |
$15–25 |
1M | claude.ai free tier | Long docs / Writing / Coding |
DeepSeek DeepSeek chat.deepseek.com ↗ |
DeepSeek V4 | $0.3–1.32 |
$1.2–3.96 |
1M | Free web app | Value / Reasoning / Chinese |
Grok xAI grok.com ↗ |
Grok 4 / 4.6 | $2–3 |
$6–15 |
256K–500K | Free tier | Real-time / Social analysis |
Qwen Alibaba Cloud qwen.ai ↗ |
Qwen3.8-Max | $1.65 |
$4.95 |
1M | Free web app | Chinese / Reasoning / Open source |
Kimi Moonshot AI kimi.com ↗ |
Kimi K2.6 / K3 | $0.95 |
$4.00 |
256K–1M | Free tier | Long context / Chinese / File analysis |
Llama Meta llama.com ↗ |
Llama 4 (Scout / Maverick) | $0.08–0.27 |
$0.30–0.85 |
1M–10M | Free open weights | Open source / Self-hosting / Enterprise |
Mistral Mistral AI mistral.ai ↗ |
Mistral Large 3 | $0.50 |
$1.50 |
256K | Le Chat free tier | EU compliance / Value / Open source |
GLM Zhipu AI zhipuai.cn ↗ |
GLM-5.3 | $1.40 |
$4.40 |
1M | Free web version | Chinese / Agent / Coding |
ChatGPT (GPT-5.6) and Claude have the strongest coding reputations; Claude's long context suits large codebases, while ChatGPT's Codex mode excels at autonomous coding. On a budget, DeepSeek's reasoning is also top-tier.
DeepSeek excels at Chinese reasoning and is free; ChatGPT and Claude also handle Chinese at native level. For daily Chinese Q&A all three work well — try before you pay for deep reasoning or long-document work.
For daily Q&A, writing, translation and light coding, the free tiers of ChatGPT / Gemini / Claude / Grok all work; DeepSeek's web app is fully free. Pay only for heavy use, strong reasoning, long documents or image generation.
APIs bill per token (1 token ≈ 0.75 English words). Input (your prompt + context) and output (generated text) are priced separately; table prices are per 1M tokens.
Data is as of Sep 2026, from official model pages and pricing pages, with links to verify. Models iterate fast — prices and versions change, and this site is maintained continuously.
This is an independent personal project by Kael Zhang, unaffiliated with OpenAI, Google, Anthropic, DeepSeek or xAI. All prices are public official information.
Four major options: Meta Llama 4 (Scout/Maverick — free weights, no license fee under 700M MAU), Alibaba Qwen (Qwen3.8-Max, free open weights), Mistral Large 3 (Apache 2.0) and DeepSeek V4 (open source). Self-hosting only costs you compute — ideal for sensitive data or high sustained volume; otherwise the hosted APIs are more convenient.