Best AI coding subscriptions under $20.

Run open-source LLMs locally or in the cloud. Ollama now offers an online coding plan with cloud access to open-source models (Llama 4, Mistral, etc.). A solid alternative to OpenC...

Google's autonomous coding agent. Jules connects to your GitHub repositories, clones them to a Cloud VM, plans changes with Gemini 3 Pro, shows a diff for review and opens a PR. It...

Sourcegraph's frontier coding agent for the terminal and editor. Amp is multi-model (GPT-5.5, Opus 4.8 and fast models) with three modes - deep, smart and rush - plus an Oracle sec...

Cognition's AI software engineer. Devin plans and executes complex engineering tasks - code migrations, PR review with visual QA, documentation, issue triage and bug fixing - and c...

The model gateway for AI apps. OpenRouter gives one OpenAI-compatible API to 400+ models across 70+ providers with auto-routing, fallback and prompt caching. The Free plan covers 2...

Fast inference for open models on LPU hardware. Groq runs Llama 4 Scout at 594 TPS, Llama 3.3 70B at 394 TPS and Llama 3.1 8B Instant at 840 TPS, with linear per-million-token pric...

European frontier AI lab. Mistral's Vibe agent runs long-horizon work and all-day coding in the CLI (Devstral), IDE (VS Code, JetBrains) and on the web, with remote coding agents a...

OpenAI's flagship with GPT-5.5 and Codex agent (released April 23, 2026). The Codex app is getting improved and the model has much more generous limits than Claude. Appreciated for...

Terminal-based agentic IDE that connects to diverse LLM providers without vendor lock-in. Models are guaranteed without quality degradation - you always get the full model. The Go ...

Microsoft's IDE extension with multi-model autocomplete built directly into VS Code and JetBrains. Now includes Opus 4.7, GPT-5.5, Gemini 3 Pro, and Grok Code. Pro+ adds GitHub Spa...

Best value for money with a smaller model and therefore generous usage limits. Ultra-fast MoE model at 100 TPS, scoring 80.2% on SWE-Bench Verified. The M2.5 variant (229B MoE) is ...

Currently the preferred open source model for coding. GLM-5.1 (released April 7, 2026) is a 754B params MoE model scoring 77.8% on SWE-Bench Verified with Code Arena Elo 1530. Stro...

AI infrastructure layer providing bonus credits at exact provider rates for 500+ models via the Kilo Gateway. The momentum reward system increases value for long-term subscribers. ...

The open source model closest to Claude in coding quality. Kimi K2.6 (released ~April 20, 2026) is a 1T params MoE with 32B active parameters, 256K context, and 4000+ tool calls pe...

Best choice for large codebases. The Context Engine indexes entire repos for deep understanding across monorepos and enterprise-scale projects. Intent workspace coordinates multipl...

Anthropic's developer-centric AI with Claude Code and Claude Design for UI work. Best-in-class models for coding, but Pro plan usage is getting very limited. Max plans are pure usa...

Market-leading AI IDE featuring Cursor 3 with Composer 2 (200+ tok/s, 61.3 on CursorBench). Multi-repo workspaces, cloud agents, agents that control their own computer, now also on...

Owned by Cognition AI. Windsurf uses daily/weekly refreshing usage quotas and features Cascade 2 with persistent Memories and Turbo Mode. SWE-1.5 model (released April 14, 2026) ru...