Ante vs Letta Code

Feature	Ante	Letta Code
Tagline	Self-contained Rust agent harness with native offline inference	The memory-first coding agent
Website	https://ante.run/	https://letta.com
Language	rust	typescript
Release Date	December 23, 2025	December 16, 2025
Last Updated	September 4, 2026	August 19, 2026
Provider Type	Independent	Independent
MCP Support	Yes	Yes
Multi-Agent Support	Yes	Yes
Git Integration	basic	basic
Multi-File Editing	Yes	Yes
Auto Test Running	No	Yes
Sandbox / Permissions	Yes	Yes
Image / Multimodal Input	Yes	Yes
Voice Mode	No	No
IDE Integration	None shipped; `ante serve` exposes a JSONL protocol for editor plugins	None - Terminal only
Hooks / Extensions	No	Yes
Web Search	Yes	Yes
Local Model Support	Yes	Yes
Context Window	Model-dependent	Model-dependent
Terminal Bench	#23 · 69.4% ±2.1 (Terminal-Bench 2.0) 1	#44 · 59.1% ±2.4 (Terminal-Bench 2.0) 2
Supported Models	17 built-in presets: Anthropic, OpenAI, Google Gemini and Vertex, xAI Grok, DeepSeek, OpenRouter, Zai, Ali Coding Plan, Antix, plus any OpenAI-compatible endpoint via ~/.ante/catalog.json and any local GGUF model through the bundled llama.cpp	Multi-provider with Claude, GPT, Gemini and self-hosted (BYOK).
Supported Subscriptions	Claude Code ChatGPT	ChatGPT GitHub Copilot Claude Code OpenCode Go / Zen
Pricing Model	Free, BYOK	Freemium
Free Tier	Yes	Yes
Price Range	Free (alpha preview)	Free (BYOK) / $20-200/mo
Source Code Availability	Proprietary Software	Open Source (Apache-2.0)
Source Code	https://github.com/AntigmaLabs/ante	c
Development State	active	active

¹ Ante tested with Gemini 3 Pro on Terminal-Bench 2.0.

² Letta Code tested with Claude Opus 4.5 on Terminal-Bench 2.0.