OpenCodevsClaude Code

We pointed both agents at Claude Opus 5.5 and captured what each sends before you type a word. Plus plans, prices, adoption, and features, checked against both vendors’ docs.

Updated October 8, 2026

Same model, different harness

Every request an agent makes carries its system prompt and tool definitions. We captured the first request from each agent on the same model to see how much that is.

System promptTool definitionsInjected context
  1. OpenCode10,468 tokens

    system prompt 2,917 · tool definitions 7,551 · 10 tools

  2. Claude Code29,220 tokens

    system prompt 1,965 · tool definitions 23,907 · injected context 3,348 · 23 tools

Tokens in the first request each agent sends for a one-word prompt, both on Claude Opus 5.5. Injected context is the environment and reminder text Claude Code adds to the first user message.[1][3]
  • Claude Code sends 2.8× as many tokens before your first message: 29,220 against 10,468.[1][3]
  • Tool definitions are most of the gap. Claude Code sends 23 tools, OpenCode sends 10.[1]
  • In dollars it’s small once cached. On Opus 5.5 the overhead costs $0.146 against $0.052 on the first turn, then $0.0058 against $0.0021 per turn read from cache.[1][2]
  • OpenCode also sends a separate 788-token request to a small Claude model to title each new session.[1]
Tools each agent sends in its first request
JobOpenCode10 toolsClaude Code23 tools
Run shell commandsbashBash
Read, edit, write filesread edit writeRead Edit Write
Find files and search codeglob grepNot in the request
Fetch a web pagewebfetchWebFetch
Search the webNot in the requestWebSearch
Delegate to subagentstaskAgent SendMessage ListAgents TaskStop
Load skillsskillSkill
Track a to-do listtodowriteNot in the request
Schedule and waitNot in the requestCronCreate CronDelete CronList ScheduleWakeup Monitor
Git worktreesNot in the requestEnterWorktree ExitWorktree
Everything elseNot in the requestNotebookEdit Workflow ReportFindings PushNotification DesignSync
  • Headless runs (claude -p and opencode run, Claude Code 2.1.291 and OpenCode 1.18.34) in an empty git repository with default settings and no MCP servers, skills, or instruction files. Each of those adds tokens in both agents.[1]
  • Tokens were counted with Anthropic’s token counting endpoint against Opus 5.5. Interactive sessions can send a different tool set.[3]
  • This measures what each agent sends, not how well it codes. Task tokens, not fixed overhead, drive most of a session’s cost.[1]

What a month costs

Set your team size and how many agent sessions each person runs a month. Each per-token option is priced on the same session, and flat plans sit alongside.

Anthropic models · either agentOpen models · OpenCode onlyFlat plan, per person

For 1 person at 200 sessions each a month, the cheapest option is DeepSeek V4 Flash at $6.99.

  1. DeepSeek V4 Flash[6]Per token

    $6.99/mo

  2. OpenCode Go[7]Flat, with caps

    $10/mo

  3. Claude Pro[23]Flat, with caps

    $20/mo

  4. OpenCode Go Plus[7]Flat, with caps

    $40/mo

  5. Kimi K2.6[6]Per token

    $44/mo

  6. GLM 5.3[6]Per token

    $68/mo

  7. Claude Sonnet 5.5[2][6]Per token

    $74/mo

  8. Claude Max 5x[24]Flat, with caps

    $100/mo

  9. Claude Opus 5.5[2][6]Per token

    $110/mo

  10. Claude Max 20x[24]Flat, with caps

    $200/mo
Claude models run in either agent. Only OpenCode can use the open models and Go; only Claude Code can use Claude plans.
  • One session is Anthropic’s published example: 1.2k input, 5.3k output, 940k cache-read, and 50k cache-write tokens. Real sessions vary, and other models may use more or fewer tokens for the same work.[25]
  • Flat plans are priced per person and have usage caps that aren’t published in tokens, so a plan may not cover a given number of sessions. Claude plans only work in Claude Code; Team standard needs at least 2 seats. OpenCode Go only serves open models.[24][23][7][18]
  • Prices per million tokens from Anthropic and OpenCode Zen. Where Zen lists no cache-write price, cache writes are counted at the input rate.[2][6]

Adoption

OpenCode is growing much faster from a smaller base. Claude Code still has far more npm downloads.

OpenCode

40.8×

npm downloads, Sep 2026 vs Oct 2025: 9.8M a month, up from 240.1K.[34]

Oct 2025Sep 2026

212,172 GitHub stars[4]

Claude Code

2.3×

npm downloads, Sep 2026 vs Oct 2025: 55.9M a month, up from 24M.[35]

Oct 2025Sep 2026

149,692 GitHub stars[36]

npm counts include CI installs and updates, and miss Claude Code’s native installer and OpenCode’s install script and Homebrew. The anthropics/claude-code repository holds plugins and issues, not the agent’s source. Stars checked October 8, 2026.

Features

Both edit files, run commands, and support MCP, skills, subagents, and custom agents.

SupportedPartial or with caveatsNot supported

Openness and models

Open-source agent[4][33]

OpenCodeMIT

Claude CodeProprietary

Model choice[5][26][19]

OpenCode75+ providers and local models

Claude CodeClaude models only

Claude Pro or Max sign-in[5][18]

OpenCodeNot allowed

Claude CodeSupported

Other subscriptions[5][18]

OpenCodeChatGPT Plus, Copilot, GitLab Duo

Claude CodeClaude plans only

Instruction files[11][29]

OpenCodeAGENTS.md; CLAUDE.md fallback in 1.x

Claude CodeCLAUDE.md; AGENTS.md fallback

Safety defaults

Approval rules[10][27]

OpenCodeAllow, ask, deny per tool and agent

Claude CodePermission modes

OS sandbox[10][28]

OpenCodeNone documented

Claude CodeBuilt in, off by default

Interfaces and tools

Desktop app[4][17]

OpenCodeBeta

Claude CodeDesktop app

Editors[14][17]

OpenCodeVS Code family, Zed, JetBrains

Claude CodeVS Code, JetBrains

Phone[17]

OpenCodeNot documented

Claude CodeClaude iOS and Android apps

Subagents[9][20]

OpenCodeGeneral, Explore, Scout + custom

Claude CodeExplore, Plan, general + custom

Language servers[12][22]

OpenCode34 built in

Claude CodeNeeds a plugin

MCP[13][21]

OpenCodeLocal and remote

Claude CodeLocal and remote

Hosting and automation

Vendor-hosted sessions[15][30]

OpenCodeNone

Claude CodeCloud sessions

Scheduled runs[31]

OpenCodeNot documented

Claude CodeRoutines (research preview)

GitHub[16][32]

OpenCode/opencode on your runners

Claude Code@claude via GitHub Actions

Run both on the same task

The overhead above says nothing about which agent writes the better patch on your code. The only way to know is to run both.

  1. 01

    Share one instruction file

    Put your conventions in AGENTS.md. OpenCode reads it natively, and Claude Code reads it when there’s no CLAUDE.md.[11][29]

  2. 02

    Use the same model

    Point OpenCode at Claude through an API key so the harness is the only difference.[5]

  3. 03

    Start from the same state

    Same commit, same dependencies, clean environment, the same three to five closed issues with known-good fixes.

  4. 04

    Compare the pull requests

    Judge the diffs, the tests each one ran, and how long review took.

Common questions

Sources and methodology

The token overhead is our own measurement. We ran each agent headless against a local endpoint that recorded its first request instead of forwarding it, then counted the parts with Anthropic’s token counting endpoint. Everything else comes from vendor documentation, pricing pages, npm, and GitHub, checked on October 8, 2026.

OpenCode has two release lines: 1.x (opencode-ai on npm) and 2.x (@opencode/cli). Rows describe 1.x unless they say otherwise. Prices and plans change often, so confirm with the vendor before you buy.

Replicas publishes this comparison and supports both harnesses. If you spot an error, email founders@replicas.dev and we'll fix it.

  1. 1Replicas measurement: first request captured from each agent · Replicas
  2. 2Claude API pricing · Anthropic
  3. 3Token counting · Anthropic
  4. 4anomalyco/opencode repository and license · GitHub
  5. 5OpenCode providers · OpenCode
  6. 6OpenCode Zen · OpenCode
  7. 7OpenCode Go · OpenCode
  8. 8OpenCode Enterprise · OpenCode
  9. 9OpenCode agents · OpenCode
  10. 10OpenCode permissions · OpenCode
  11. 11OpenCode rules (AGENTS.md) · OpenCode
  12. 12OpenCode LSP servers · OpenCode
  13. 13OpenCode MCP servers · OpenCode
  14. 14OpenCode IDE and ACP support · OpenCode
  15. 15OpenCode server · OpenCode
  16. 16OpenCode GitHub integration · OpenCode
  17. 17Claude Code overview · Anthropic
  18. 18Claude Code legal and compliance · Anthropic
  19. 19Claude Code LLM gateway configuration · Anthropic
  20. 20Claude Code subagents · Anthropic
  21. 21Claude Code MCP · Anthropic
  22. 22Claude Code tools reference · Anthropic
  23. 23Claude plans and pricing · Anthropic
  24. 24Claude Max plan tiers and pricing · Anthropic
  25. 25Manage Claude Code costs · Anthropic
  26. 26Claude Code on third-party platforms · Anthropic
  27. 27Claude Code permission modes · Anthropic
  28. 28Claude Code sandboxing · Anthropic
  29. 29Claude Code memory and CLAUDE.md · Anthropic
  30. 30Use Claude Code in the cloud · Anthropic
  31. 31Claude Code routines · Anthropic
  32. 32Claude Code GitHub Actions · Anthropic
  33. 33anthropics/claude-code license · GitHub
  34. 34opencode-ai download counts · npm
  35. 35@anthropic-ai/claude-code download counts · npm
  36. 36anthropics/claude-code repository · GitHub