AI model equivalents

AI model equivalents: the Opus, Sonnet and Haiku of every coding CLI

You know the Claude lineup by heart, then you open Codex, Gemini or Grok and none of the names tell you which model plays the role of Opus. This page places every model of every coding agent CLI AgentsRoom runs on one scale, so each one reads in terms you already know.

Read live from the AgentsRoom model catalogue. Popularity as of Oct 6, 2026, 4:16 PM.

CLIs compared

12

Models placed

94

Launches counted, last 30 days

51,051

The short answer

Claude is the yardstick because it is the lineup most developers know. Each line names the models of other vendors that play the same role today.

Fable class

Fable-class (frontier) equivalents

The frontier models, one step above Opus: GPT-6 Astra in Codex, Daybreak Blue in Codex, and GPT-5.6 Sol Max in Codex.

For the hardest problems, at the highest price.

Opus class

Opus equivalents

The models that play the role of Claude Opus: GPT-6.1 Sol in Codex, Gemini 3.1 Pro (High) in Antigravity, and Grok 4.7 in Grok.

The flagship most serious coding work runs on.

Sonnet class

Sonnet equivalents

The models that play the role of Claude Sonnet: GPT-6 Luna in Codex, GPT-5.6 Terra in Codex, Gemini 3.8 Flash (High) in Antigravity, GLM 5.3 in OpenCode, Qwen3.8 Max in OpenCode, DeepSeek in Aider, Medium 3.5 in Mistral Vibe, and Kimi K3 in Kimi.

Balanced: strong enough for most features, cheaper than the flagship.

Haiku class

Haiku equivalents

The models that play the role of Claude Haiku: Codex Spark in Codex, Qwen3 Coder in Ollama, GLM 4.6 in Ollama, and DeepSeek V4 Flash in oh-my-pi.

Light and fast: renames, small fixes, summaries.

Every coding CLI, tier by tier

One row per CLI AgentsRoom runs, one column per tier. Current models come first; the older versions a list still offers are counted under them.

AI model equivalents by coding agent CLI and capability tier
CLIFrontierFable classOpus classSonnet classHaiku class
Claude
Fable
OpusOpus 5.5
+2 older versions

Opus 5, Opus 4.8

SonnetSonnet 5.5
+1 older version

Sonnet 5

Haiku
Codex
GPT-6 AstraDaybreak BlueDaybreak RedGPT-5.6 Sol Max
GPT-6.1 Sol
+3 older versions

GPT-6 Sol, GPT-5.6 Sol, GPT-5.5

GPT-6 LunaGPT-5.6 Terra
+4 older versions

GPT-5.6 Luna, GPT-5, GPT-5.4, GPT-5.2

Codex Spark
+1 older version

GPT-5.4 Mini

AntigravityNone
Gemini 3.1 Pro (High)Claude Opus 5.5 (High)Claude Opus 5.5 (Medium)
Gemini 3.8 Flash (High)Gemini 3.8 Flash (Medium)Claude Sonnet 5.5 (High)Claude Sonnet 5.5 (Medium)
+5 older versions

Gemini 3.7 Flash (High), Gemini 3.7 Flash (Medium), Gemini 3.6 Flash (High), Gemini 3.6 Flash (Medium), Gemini 3.5 Flash (Medium)

None
OpenCodeNone
Claude Opus
Claude SonnetGPT-5Kimi K3GLM 5.3Qwen3.8 Max
None
AiderNone
Claude Opus
Claude SonnetDeepSeek
GPT-4o
GrokNone
Grok 4.7
NoneNone
Mistral VibeNoneNone
Medium 3.5
None
OllamaNoneNone
Qwen3 Coder 480B
Gemma 3Qwen3 CoderGLM 4.6
GitHub Copilot
GPT-6 Astra
GPT-6.1 SolClaude Opus 5.5GPT-5.5
+5 older versions

GPT-6 Sol, Claude Opus 5, GPT-5.6 Sol, Claude Opus 4.8, Claude Opus 4.7

Claude Sonnet 5.5GPT-6 LunaGPT-5.6 TerraGemini 3.7 FlashKimi K3
+4 older versions

Claude Sonnet 5, GPT-5.6 Luna, Claude Sonnet 4.6, Gemini 3.6 Flash

Claude Haiku 4.5MAI-Code-1.1-Flash
KimiNoneNone
Kimi K3Kimi K3-256KKimi K2.8 PreviewK2.7 Code HighSpeed
None
oh-my-piNone
Claude Opus 5.5GPT-6 Sol
Claude Sonnet 5Gemini FlashGLM 5.3
+1 older version

GLM 5.2

DeepSeek V4 Flash
DevinNone
Claude OpusGPTCodexGemini
SWEClaude SonnetGLMKimiDeepSeek
Claude Haiku
AmpPicks the model for you on every request: no fixed tier.
FreebuffPicks the model for you on every request: no fixed tier.
CursorPicks the model for you on every request: no fixed tier.
  • Among the three models of its CLI the community launches most
  • Older version: a newer generation of the same model is in the same list

Same model, several CLIs

A model family is not tied to its vendor's tool. Here is every family that more than one CLI ships today, with the name each CLI gives it.

Claude Opus

Opus class

7 CLIs

  • ClaudeOpus 5.5
  • AntigravityClaude Opus 5.5 (High)
  • OpenCodeClaude Opus
  • AiderClaude Opus
  • GitHub CopilotClaude Opus 5.5
  • oh-my-piClaude Opus 5.5
  • DevinClaude Opus

Claude Sonnet

Sonnet class

7 CLIs

  • ClaudeSonnet 5.5
  • AntigravityClaude Sonnet 5.5 (High)
  • OpenCodeClaude Sonnet
  • AiderClaude Sonnet
  • GitHub CopilotClaude Sonnet 5.5
  • oh-my-piClaude Sonnet 5
  • DevinClaude Sonnet

Kimi

Sonnet class

4 CLIs

  • KimiKimi K3
  • OpenCodeKimi K3
  • GitHub CopilotKimi K3
  • DevinKimi

GLM

Sonnet class

4 CLIs

  • OpenCodeGLM 5.3
  • OllamaGLM 4.6
  • oh-my-piGLM 5.3
  • DevinGLM

Claude Haiku

Haiku class

3 CLIs

  • ClaudeHaiku
  • GitHub CopilotClaude Haiku 4.5
  • DevinClaude Haiku

GPT Sol

Opus class

3 CLIs

  • CodexGPT-6.1 Sol
  • GitHub CopilotGPT-6.1 Sol
  • oh-my-piGPT-6 Sol

Gemini Flash

Sonnet class

3 CLIs

  • AntigravityGemini 3.8 Flash (High)
  • GitHub CopilotGemini 3.7 Flash
  • oh-my-piGemini Flash

DeepSeek

Sonnet class

3 CLIs

  • AiderDeepSeek
  • oh-my-piDeepSeek V4 Flash
  • DevinDeepSeek

GPT Astra

Fable class

2 CLIs

  • CodexGPT-6 Astra
  • GitHub CopilotGPT-6 Astra

GPT Luna

Sonnet class

2 CLIs

  • CodexGPT-6 Luna
  • GitHub CopilotGPT-6 Luna

GPT Terra

Sonnet class

2 CLIs

  • CodexGPT-5.6 Terra
  • GitHub CopilotGPT-5.6 Terra

Qwen

Sonnet class

2 CLIs

  • OpenCodeQwen3.8 Max
  • OllamaQwen3 Coder

What developers actually launch

The most launched models of the last 30 days among AgentsRoom users, read in tier terms. Equivalents tell you which models play the same role; launches tell you which ones people pick.

  1. 1

    OpusOpus class

    Claude439 members

    22.9% of launches

  2. 2

    SonnetSonnet class

    Claude173 members

    12.8% of launches

  3. 3

    Opus 5.5Opus class

    Claude103 members

    9.4% of launches

  4. 4

    GPT-5.6 SolOpus class

    Codex119 members

    5% of launches

  5. 5

    GPT-6 LunaSonnet class

    Codex47 members

    4.9% of launches

Full usage statistics, by CLI and by model

How the tiers are assigned

One scale with four tiers, anchored on Claude: Fable class (frontier), Opus class (flagship), Sonnet class (balanced) and Haiku class (light). Every model family gets the tier of the role its vendor gives it in its own lineup, and the same family keeps the same tier in every CLI that ships it.

OpenAI's Astra is a frontier model, Sol its flagship, Luna and Terra its balanced models, Spark and the mini models its light ones. Gemini Pro sits in the Opus class and Gemini Flash in the Sonnet class. Grok Heavy is frontier and Grok itself is Opus class.

Open-weight families (Kimi, GLM, DeepSeek, Qwen and others) are placed in the Sonnet class at most, and their light cuts in the Haiku class. Between two tiers, the table always picks the lower one: it would rather undersell a model than tell you it can replace Opus when it has not shown it.

Routers get no tier, because they pick a model on each request: GitHub Copilot (Auto), Amp (Amp default), Freebuff (Freebuff default), Devin (Adaptive), and Cursor (Default). Neither does a family the table does not know yet: no equivalence is better than a wrong one.

A tier is not a benchmark score. Two models of the same tier still differ in price, speed, context window, usage limits and the tasks they do best. Use the table to know where to look, then try both on your own task.

Data: the model catalogue AgentsRoom serves to its desktop and mobile apps, read at every visit, and the tier rules the app applies under every model list. Popularity: real agent launches by AgentsRoom users who share usage data, over the last 30 days, refreshed every ten minutes.

Frequently asked questions

Which Codex model is the equivalent of Claude Opus?

GPT-6.1 Sol. In the table above, Claude Opus and GPT-6.1 Sol both sit in the Opus class: the flagship most serious coding work runs on, one step below the frontier tier. Codex's frontier model, the counterpart of Fable, is GPT-6 Astra. For Sonnet-class work, the Codex model is GPT-6 Luna. And the Haiku-class model of the list is Codex Spark.

What is the Gemini equivalent of Claude Opus?

Gemini 3.1 Pro (High), in Antigravity. Gemini Pro is Google's flagship line, so it sits in the Opus class. Gemini Flash plays the role of Sonnet: today Gemini 3.8 Flash (High).

Which GPT model is the equivalent of Claude Sonnet?

GPT-6 Luna, in Codex. OpenAI's balanced line sits in the Sonnet class: strong enough for most features, cheaper than the flagship.

What is the OpenAI equivalent of Claude Haiku?

Codex Spark, in Codex: the light, fast model of the lineup, for renames, small fixes and quick questions.

Is GPT-6.1 Sol as good as Opus 5.5?

Same tier does not mean same results. GPT-6.1 Sol and Opus 5.5 play the same role in their vendor's lineup, but they still differ in price, speed, context window, usage limits and the tasks they handle best, and benchmark rankings between the two change with every release. The reliable answer is to run both on the same task of yours: AgentsRoom launches one agent on each, side by side, in the same project.

Where can I use Claude Opus outside Claude Code?

Today, Opus is also in the model lists of Antigravity, OpenCode, Aider, GitHub Copilot, oh-my-pi, and Devin, under the name each CLI gives it. Same family, same tier: the equivalence does not change with the CLI that serves it.

Why does the table cap Kimi, GLM or Qwen at Sonnet class?

Because the scale follows the role a model plays, and it never oversells. Open-weight families are often excellent value and several CLIs ship them, but placing one in the Opus class would tell you it can replace Opus on any task, which none has shown across the board. Their light cuts sit in the Haiku class.

Why do Auto and Default have no equivalent?

Because they are not a model: GitHub Copilot (Auto), Amp (Amp default), Freebuff (Freebuff default), Devin (Adaptive), and Cursor (Default) let the service pick a model on each request, so there is no fixed capability to compare. The table leaves them out rather than guess.

How up to date is this table?

It is computed every time the page is opened, from the same model catalogue the AgentsRoom app serves to every desktop and phone. When AgentsRoom adds a model a CLI has released, it appears here at once, in its tier, and the version it replaces is marked older. The popularity flames are refreshed every ten minutes from the last 30 days of launches.

Can AgentsRoom pick the right model for my task?

Yes, in two ways, and both only suggest. AI Suggestion sizes the task before the session opens (Quick, Standard, Complex or Extreme) and opens the agent on the cheapest model of its CLI that can still do it. Adaptive Mode reads your draft once the CLI is running and suggests a lighter model when the task does not need the flagship, or tells you the current one is right.

In AgentsRoom

AgentsRoom answers this inside your model picker

AgentsRoom is a desktop command center for AI coding agents. Claude, Codex, GitHub Copilot CLI, Cursor and 10 other agent CLIs run side by side, each agent on the model you choose. The equivalences of this page are the ones the app shows you while you choose, and a few smart features help you size the model to the task.

≈ Opus under every model

Every model list (new agent, quick launch, backlog ticket, mobile app) tags each model with its counterpart in the CLI you know best, and sorts by power, price or popularity.

Model equivalents in the app

Let the task pick the model

AI Suggestion turns a one-line description into a complexity level (Quick, Standard, Complex, Extreme) and opens the agent on the cheapest model that fits. Adaptive Mode checks your draft and suggests a lighter model when the flagship is overkill.

AI Suggestion and Adaptive Mode

Overflow to the equivalent model

When an account nears its limit, a quota rule can start new agents on another CLI model by model: Fable, Opus and Sonnet each go to their counterpart of the same tier, pre-filled from these tiers.

Quota rules and usage alerts

The right model per ticket

Pin a provider, a model and a reasoning effort on a backlog ticket, so the agent that runs it uses exactly that, and keep the flagship for the tickets that need it.

Backlog task board

Hand the tests to a cheaper model

A developer agent on a flagship model hands the test run to a QA agent on a lighter one over MCP, so the expensive model stays on the code.

Agent delegation

A model per step

In an agent team, every step has its own role, CLI and model: the architect on a frontier model, the implementer on the Opus class, the reviewer wherever it pays off.

Agent teams
Download AgentsRoom

Further reading

Claude Code vs Codex CLI: which one to pick in 2026 (and why not both)

An honest comparison of Claude Code and Codex CLI: pricing, speed, MCP, control. And why, after a few weeks, I stopped choosing between them.

AgentsRoom Now Supports Ollama: Run Local Models Next to the Cloud

Ollama is now a provider in AgentsRoom. Run local open-source models like Llama, Qwen, Gemma and DeepSeek next to cloud agents, and set a local-or-cloud dial on each agent, switchable mid-conversation.