One model ladder: Claude and GPT
Published Updated
This is the Learning Atlas from Orbit. Get your invite
AI providers and modelsAI agents and harnesses
-
Step 1: Anthropic
Why this matters here
Start with Anthropic so the learner can place the Claude family with its provider before comparing capability tiers.
In short
An AI safety and research company that develops the Claude family of large language models. They focus on building helpful, harmless, and honest AI systems.
-
Step 2: OpenAI
Why this matters here
OpenAI adds the second provider, making clear that the ladder compares equivalent roles across two model families.
In short
A leading AI research and deployment company that created the GPT family of models and ChatGPT. They provide highly capable generalized models via API.
-
Step 3: Claude (model family)
Why this matters here
Claude groups Anthropic's tiers under one family before the individual capability steps begin.
In short
A family of large language models developed by Anthropic, heavily utilized for coding, agentic orchestration, and complex analysis due to their deep reasoning capabilities.
-
Step 4: GPT (model family)
Why this matters here
GPT provides the parallel OpenAI family so the learner can follow its tiers on the same capability ladder.
In short
The foundational series of large language models created by OpenAI, driving ChatGPT and widely integrated into developer tools and enterprise APIs.
-
Step 5: Haiku
Why this matters here
Haiku starts the capability sequence as the lightweight Claude tier for routine work.
In short
The fastest, most compact model in the Claude family, designed for near-instant execution of high-volume, low-latency tasks.
-
Step 6: Luna
Why this matters here
Luna shares the lightweight rung, showing the comparable GPT tier alongside Haiku.
In short
OpenAI's small GPT tier, suited to quick, routine tasks where responsiveness and efficient use of a token budget matter most.
-
Step 7: Sonnet
Why this matters here
Sonnet raises the ladder to a more capable Claude tier for work that needs a stronger balance of speed and depth.
In short
The balanced middle tier of the Claude model family. It is widely considered the sweet spot for agentic coding and general enterprise workloads.
-
Step 8: Sol
Why this matters here
Sol is the corresponding GPT step, keeping the comparison focused on capability rather than provider preference.
In short
OpenAI's middle GPT tier, balancing capability and speed for coding, analysis, and agent work that is more involved than routine requests.
-
Step 9: Opus
Why this matters here
Opus introduces the next higher Claude tier for difficult work that needs deeper reasoning.
In short
A high-capability model in the Claude family, one tier below Fable, designed to handle highly complex, cognitively demanding tasks.
-
Step 10: Astra
Why this matters here
Astra supplies the matching higher GPT tier before the ladder reaches its highest Claude tier.
In short
OpenAI's frontier GPT tier, intended for the most demanding coding, analysis, and multi-step reasoning work in the GPT ladder.
-
Step 11: Fable
Why this matters here
Fable completes the capability ladder as the highest Claude tier in this comparison.
In short
The frontier-class model in the Claude family, representing the highest tier of intelligence and reasoning capability.
-
Step 12: Claude Code
Why this matters here
Claude Code shows how a Claude model becomes a coding agent that can work with project files and tools.
In short
An autonomous agentic coding assistant built by Anthropic that operates directly within the terminal to read codebases, edit files, and execute commands.
-
Step 13: Codex CLI
Why this matters here
Codex CLI completes the path with the corresponding terminal coding-agent experience for GPT models.
In short
An official terminal-based AI coding agent from OpenAI that reads local repositories, edits files, and executes commands autonomously.