AI coding assistants, code completion models, and developer-focused AI tools.
AI pair programmer integrated into every major IDE
Industry standard for AI-assisted coding. Model picker spans GPT-5, Claude, and Gemini for completions, chat, and agent mode for autonomous task completion. Works across 20+ languages with deep IDE integration.
AI-first code editor built on VS Code with multi-model support
Redefined AI coding with Composer mode, multi-file editing, and codebase-aware context. Supports Claude, GPT-5, and custom models. Go-to for power users wanting maximum AI capability.
Anthropic's agentic coding tool for terminal-based development
Agentic coding from the terminal with full repo understanding. Excels at multi-file refactors, test writing, and complex architectural changes. Best for developers who prefer CLI workflows.
Top open-source code model with 236B MoE architecture
Best open-weight code model. Matches proprietary models on HumanEval and SWE-bench. Cost-effective for self-hosted code generation and completion at scale.
Alibaba's code-specialised model with 80B parameters
Strong code generation especially for Python and JavaScript. Excellent multilingual code comments and documentation. Self-hostable on 2× A100 GPUs.
Mistral's 22B code model optimised for low-latency FIM
Excellent fill-in-the-middle completions with very low latency. Best for IDE integration where speed matters. 32K context window covers most files. EU data residency available.
AI-native IDE (formerly Codeium), now part of Cognition, with Cascade agentic flows
Strong Copilot alternative with Cascade for multi-step agentic coding. Acquired by Cognition (makers of Devin) in mid-2025 after a bidding process; development continues under that roof. Good for teams wanting an all-in-one AI IDE without per-seat Copilot costs.
Meta's code-specialised Llama variant for generation and infilling — no longer the current choice
Superseded by DeepSeek Coder V3 and newer open code models. Meta has not shipped a direct successor, and the ecosystem has moved on. Fine for legacy fine-tunes already built on it; do not start new work here.
BigCode open-source code model trained on The Stack v2
Truly open model with transparent training data provenance. Good baseline for fine-tuning on proprietary codebases. 15B variant runs on consumer GPUs.
AWS's AI coding assistant with deep cloud integration
Best for AWS-heavy teams. Code transformation (Java 8→17) is a killer feature. Inline completions decent but behind Copilot. Free tier available.
Privacy-first AI code assistant with on-premise deployment
Best for enterprises requiring complete code privacy — models run fully on-prem. Completion quality behind Copilot but improving. Strong compliance story for regulated industries.
In-browser AI coding with instant deployment
Unique proposition: code, run, and deploy entirely in the browser with AI. Replit Agent matured well past its early rough edges and now handles substantial full-stack builds from a prompt. Best for rapid prototyping and small production apps; less suited for large enterprise codebases.
