Price, features, supported models … We compared the three best “AI Native” ideas on the market today.
The race for AI development environments is intensifying. Amazon in turn unveiled in July its own agentic ideas. Based on Claude models and with attractive pricing, Kiro is an interesting alternative to Cursor and Windsurf. Functionities, models available, pricing… Here are the main differences between the three behemoths of the code.
Very similar features
| Functionality | Cursor | Kiro | Windsurf |
|---|---|---|---|
| Code completion | ✓ | ✓ | ✓ |
| Multi-Fichiers edition | ✓ | ✓ | ✓ |
| Agent mode (autonomous) | ✓ | ✓ | ✓ |
| Generation of terminal orders | ✓ | ✓ | ✓ |
| Debugging | ✓ | ✓ | ✓ |
| Web search | ✓ | ✗ | ✓ |
| MCP support | ✓ | ✓ | ✓ |
| Generation of architecture diagrams | ✗ | ✓ | ✗ |
| Automated code review | ✓ | ✓ | ✓ |
| “Spec-Driven” agent | ✗ | ✓ | ✗ |
| Automatic documentation update | ✓ | ✓ | ✓ |
The differences are nestled in detail: Amazon Kiro ignores web research, a significant limitation which prevents its agent from automatically seeking online documentation or solving problems based on external resources. In return, Amazon’s IDE is distinguished by its native capacity to generate architecture diagrams and by its Spec-Driven mode. The latter automatically transforms requests into detailed technical specifications before having them executed by several agents unlike the free conversational mode of other FDI.
Model availability: limited kiro, large winner cursor
| Model | Cursor | Kiro | Windsurf |
|---|---|---|---|
| Claude 3 opus | ✓ | ✗ | ✗ |
| Claude 3.5 Haiku | ✓ | ✗ | ✗ |
| Claude 3.5 SONNET | ✓ | ✗ | ✓ |
| Claude 3.7 SONNET | ✓ | ✓ | ✓ |
| Claude 4 opus | ✓ | ✗ | ✓ |
| Claude 4 SONNET | ✓ | ✓ | ✓ |
| Cursor Small | ✓ | ✗ | ✗ |
| Deepseek R1 | ✓ | ✗ | ✓ |
| Deepseek V3 | ✓ | ✗ | ✗ |
| Gemini 2.0 Flash | ✗ | ✗ | ✓ |
| Gemini 2.0 Pro (EXP) | ✓ | ✗ | ✗ |
| Gemini 2.5 Flash | ✓ | ✗ | ✓ |
| Gemini 2.5 Pro | ✓ | ✗ | ✓ |
| GPT 4.1 | ✓ | ✗ | ✓ |
| GPT-5 | ✓ | ✗ | ✓ |
| GPT-5 Fast | ✓ | ✗ | ✓ |
| GPT-5 Mini | ✓ | ✗ | ✓ |
| GPT-5 NANO | ✓ | ✗ | ✓ |
| Grok Fast Code | ✓ | ✗ | ✓ |
| GPT-4O | ✓ | ✗ | ✓ |
| GPT-4O Mini | ✓ | ✗ | ✗ |
| GROK 2 | ✓ | ✗ | ✗ |
| GROK 3 | ✓ | ✗ | ✓ |
| GROK 3 Mini | ✓ | ✗ | ✓ |
| GROK 4 | ✓ | ✗ | ✗ |
| Kimi K2 | ✗ | ✗ | ✓ |
| O1 | ✓ | ✗ | ✗ |
| O1 Mini | ✓ | ✗ | ✗ |
| o3 | ✓ | ✗ | ✓ |
| O3-mini | ✓ | ✗ | ✗ |
| O4-mini | ✓ | ✗ | ✗ |
| Qwen3-Coder | ✗ | ✗ | ✓ |
| Swe-1 | ✗ | ✗ | ✓ |
| Swe-1-Lite | ✗ | ✗ | ✓ |
Amazon Kiro adopts a minimalist approach And is limited for the moment to Claude 3.7 and Claude 4 SONNET. A restriction of choice which is explained by a deliberate strategy: favoring quality to diversity. The Claude models enjoy an excellent reputation with developers for their ability to work independently on complex coding tasks. However, the Kiro team plans to support more models in the future. We can certainly wait for the support of several anthropic models (Claude 4 opus for example) and potentially an Amazon house model derived from the Nova family.
Cursor for his part is positioned as the Champion of Diversity With 21 supported models, almost integrating the catalogs of Openai, Google, Anthropic and Deepseek. The publisher is also known to quickly integrate the models a few hours after their official release.
Finally, Windsurfs occupies an intermediate position With 22 models, distinguished by its house models from the SWE family specially designed for development, as well as by the exclusivity of Kimi K2, which impressed by its performance in July.
Pricing: different strategies
Each IDE adopts a different billing system to count the use of AI models. Cursor uses “requests” corresponding to a complete interaction with the IA agent (context analysis, code generation, response). Windsurf works with “credits” whose number varies depending on the model used. For example, a request with Claude Sonnet 4 consumes approximately 2 credits when the same request with GPT-4.1 consumes 0.25. The difference is notable.
Finally, Amazon Kiro distinguishes two types of interactions: “vibe requests” for conventional conversations (explain code, correct a bug …) and “specimens” for the execution of complete tasks in Spec-Drive mode (implement an entire functionality from a specifications).
From $ 0 to 60 per month for Windsurf
On the formula side, Windsurf offers a generous free offer of 25 monthly credits and a pro plan at only $ 15 per month including 500 credits. The platform also offers Teams plans at $ 30 per user per month with 500 credits but a support and company security. Finally an Enterprise offer from $ 60 per user per month allows you to use up to 1,000 credit per month.
From $ 0 to 200 dollars per month for Cursor
Cursor adopts a premium strategy with three individual offers: free hobby but limited in the number of calls for models (without the publisher figures the limit), pro at 20 dollars per month and ultra at 200 dollars per month for intensive users. The pro plan allows approximately 225 requests with Claude 4 Sonnet per month, while the Ultra offers up to 4,500 monthly requests. Teams plans start at 40 dollars per user per month with 500 queries included per user, or 33% more expensive than Windsurf.
From $ 0 to 200 per month for Kiro
Amazon Kiro remains completely free for the moment during the preview phase. The announced plans will include a free offer with 50 vibe Requests, pro at 20 dollars per month with 225 vibe and 125 specimens, pro+ at 40 dollars per month with 450 vibe and 250 specimens, and Power at 200 dollars per month with 2,250 vibe and 1,250 specimens. Amazon therefore aligns clearly with Cursor prices.
Which ide to choose?
Windsurf remains the best compromise for the majority of developers. With 22 supported models and an attractive price at $ 15 per month, it offers a large enough palette for reasonable pricing. Its flexible credits system makes it the ideal solution for independent developers and small teams who discover the generative AI.
Cursor is for power users who seek excellence without budgetary constraints. Finally, Amazon Kiro targets a specific audience: developers who favor quality to quantity with its Spec-Driven mode. Its current free of charge makes it an opportunity to seize, even if future prices will line up on competition.
