1 · Top Score
Codex
OpenAI
OpenAI94.0
OpenAI agentic coding product spanning local CLI, cloud environments and multi-agent workflows for software development.
coding-agentRealtimeTool UseCode Exec
SWEN Agent Registry
Best AI Agents 2026Coding, Research and Voice Agents
This page launches SWEN’s canonical agent vertical. The score evaluates autonomy, reliability, tool use, value and integration for real agentic products.
5
canonical products ranked
5
with public API or developer access
5
with operational autonomy or tool execution
2 · Top Score
Claude Code
Anthropic
Anthropic93.2
Anthropic agentic coding system designed to read codebases, edit files, run commands and automate development tasks across environments.
coding-agentRealtimeTool UseCode Exec
3 · Top Score
Devin
Cognition
91.6
Cloud-based autonomous software engineer aimed at real engineering teams, multi-repo work and delegated ticket execution.
coding-agentRealtimeTool UseCode Exec
Full Ranking
Composite score built from autonomy, reliability, tool use, value and integration.
0 open source
| # | Model | Company | Score | Autonomy | Reliability | Tool Use | Price | Highlights | Release |
|---|---|---|---|---|---|---|---|---|---|
| 1 | Codex OpenAI agentic coding product spanning local CLI, cloud environments and multi-agent workflows for software development. |  OpenAI | 94.0 | 95 | 92 | 96 | Included in ChatGPT plans / platform usage | Environment: Terminal + cloud worktrees • Primary mode: Agentic coding • Tool use: Native | — |
| 2 | Claude Code Anthropic agentic coding system designed to read codebases, edit files, run commands and automate development tasks across environments. |  Anthropic | 93.2 | 94 | 93 | 94 | Usage-based through Claude platform | Environment: Terminal + IDE + browser • Primary mode: Agentic coding • Tool use: Native | — |
| 3 | Devin Cloud-based autonomous software engineer aimed at real engineering teams, multi-repo work and delegated ticket execution. | Cognition | 91.6 | 93 | 88 | 92 | Plans from US$20/month | Environment: Cloud software engineer • Primary mode: Parallel coding agents • Tool use: VM + SaaS | — |
| 4 | Deep Research OpenAI research agent optimized for browsing, analysis and synthesis across hundreds of sources rather than software engineering execution. | OpenAI | 90.4 | 92 | 90 | 91 | Available via ChatGPT / API models | Environment: Web + files + MCP • Primary mode: Research agent • Tool use: Search + file workflows | — |
| 5 | ElevenAgents Production platform for conversational voice agents with visual builder, programmatic control and native voice stack integration. | ElevenLabs | 89.8 | 88 | 90 | 86 | Startup grants / platform pricing | Environment: Voice agent platform • Primary mode: Conversational voice agents • Tool use: Native workflows | — |
Score Breakdown
Autonomytop model reference
Reliabilitytop model reference
Tool Usetop model reference
FAQ
Why create an agent registry instead of mixing agents into code rankings?
Because agents are products, not just base models. They combine models, tools, environments and orchestration. A dedicated registry lets SWEN rank autonomy, reliability, tool use and integration without pretending this is the same as ranking a raw LLM.
Can voice agents and coding agents live on the same page?
Yes. The page is agent-first. Each entry carries its own subcategory and descriptors, so the ranking can compare broad operational quality while still showing whether the product is a coding, research or voice agent.
Is this already a live automated vertical?
Yes. Once deployed, this page is refreshed by the same universal benchmark cron that updates the other canonical registries.