ChatGPT vs Claude: Which Model Matches Which (2026)

 · 6 min read  ·

You normally use Claude Opus 5.5. Today you need an OpenAI model, and you can’t remember if that’s Sol, Terra, Luna, or Astra.

Or you want to start handing the easy tasks to a cheaper model. What’s the cheap one at OpenAI, and what’s the cheap one at Claude?

Is there a table, a comparison I can just glance at? Here you go.

Flagship

OpenAI

GPT-6.2 Astra

gpt-6.2-astra

Claude

Fable 5.2

claude-fable-5-2

Hardest projects and long agent jobs. $8 / $40

Heavy daily

OpenAI

GPT-5.6 Sol

gpt-5.6-sol

Claude

Opus 5.5

claude-opus-5-5

No GPT-6.x model at this price. Opus is where Anthropic says to start. $4 / $20, promo through Nov 30

Everyday

OpenAI

GPT-6.2 Sol

gpt-6.2-sol

Claude

Sonnet 5.5

claude-sonnet-5-5

Everyday coding, writing, and agents. $2 / $10

Fast middle

OpenAI

GPT-6.1 Terra

gpt-6.1-terra

Claude

No equivalent

Screens, UI actions, and drafts. $1 / $5

Fast, cheap

OpenAI

GPT-6 Luna

gpt-6-luna

Claude

Haiku 5.5

claude-haiku-5-5

Routing, extraction, and high volume. $0.10 / $0.50, then Haiku costs more

Last updated November 20, 2026. Prices are standard API rates per million tokens, input then output, for a short prompt. I checked OpenAI’s models index and Claude’s models overview that day.

Quick answers

If you use

GPT-6.2 Astra

Closest equivalent

Claude Fable 5.2

$8 / $40 in/out per 1M

In Work and Codex, not the Chat GPT-6 button.

If you use

Claude Fable 5.2

Closest equivalent

GPT-6.2 Astra

$8 / $40 in/out per 1M

Credits on Claude Pro. Max and Ultra include it for up to half the weekly limit.

If you use

GPT-6.2 Sol

Closest equivalent

Claude Sonnet 5.5

$2 / $10 in/out per 1M

Sonnet is on every Claude plan, including Free. Sol starts at Plus.

If you use

Claude Sonnet 5.5

Closest equivalent

GPT-6.2 Sol

$2 / $10 in/out per 1M

The Chat GPT-6 button is an earlier Sol, GPT-6 Sol.

If you use

Claude Opus 5.5

Closest equivalent

GPT-5.6 Sol

$4 / $20 in/out per 1M

No GPT-6.x model at this price. Promo through Nov 30, 2026.

If you use

GPT-5.6 Sol

Closest equivalent

Claude Opus 5.5

$4 / $20 in/out per 1M

Older model. For new work, GPT-6.2 Sol is $2 / $10.

If you use

GPT-6 Luna

Closest equivalent

Claude Haiku 5.5

$0.10 / $0.50 in/out per 1M

Haiku rises to $0.50 / $2.50 past 100k prompt tokens.

If you use

Claude Haiku 5.5

Closest equivalent

GPT-6 Luna

$0.10 / $0.50 in/out per 1M

Free and Go Chat run GPT-5.6 Luna, not this one.

If you use

GPT-6.1 Terra

Closest equivalent

No Claude equivalent

$1 / $5 in/out per 1M

Screens and drafts. Sonnet is the next Claude rung, at twice the price.

The details

Pricing caveats, context, and cutoffs

Prices in the chart are short-context standard rates from OpenAI pricing and Claude pricing.

  • OpenAI context on these models is 1,050,000 tokens (922,000 max input, 128,000 max output). Claude’s four current models list a 1M window and 128,000 max output.
  • Past 272,000 input tokens, Astra, Sol, Terra, Luna, and GPT-5.6 Sol bill the whole request at 2x input and 1.5x output.
  • Haiku jumps at 100,000 prompt tokens ($0.50 / $2.50). Fable, Opus, and Sonnet stay on one rate for the full 1M window. Anthropic says a 900k-token request on those models costs the same per token as a 9k-token request.
  • Cached input, short context: Astra $0.40, GPT-6.2 Sol $0.10, Terra $0.05, Luna $0.01, GPT-5.6 Sol $0.40. Claude cache hits: Fable $0.20, Opus $0.20, Sonnet $0.10, Haiku $0.01 on the short band. Batch is half of standard on both sides.
  • Knowledge cutoffs: Astra and Luna, May 18, 2026. GPT-6.2 Sol and Terra, June 1, 2026. GPT-5.6 Sol, February 16, 2026. The four current Claude models list June 2026.
  • GPT-5.6 Sol’s $4 / $20 rate is promotional through November 30, 2026, per its model page.
Where each model shows up
  • OpenAI’s index says start at Astra. Anthropic’s overview says start at Opus 5.5, and move to Fable when Opus at a higher effort is not enough.
  • In ordinary Chat, GPT-6 Sol covers Instant through Extra High. Pro effort is GPT-6 Astra. Plus includes Medium and High. Extra High and Pro effort are on Pro, Business, and Enterprise. Free and Go use GPT-5.6 Luna, including Think.
  • Mythos 5.2 costs the same as Fable and is limited to Anthropic’s verification programs. OpenAI’s Daybreak models are limited to approved accounts. I left both off the chart.
Reasoning effort

A higher effort spends more output tokens. You stay on the same model.

  • OpenAI: reasoning.effort. The reasoning guide says GPT-6.2 Astra and Sol reject none. The GPT-6.2 guide says start at medium, and use low where an older call used none or minimal. Astra’s listed levels are low, medium, high, xhigh, and max. Luna and Terra accept none and default to medium. Reasoning tokens count as output tokens.
  • reasoning.mode can be standard or pro. Pro does more work at the same token rates. The guide’s example sets Sol to pro.
  • Claude: effort is low, medium, high, xhigh, and max. API defaults are high on Fable and Sonnet, medium on Opus and Haiku. Claude Code defaults all four to medium. You cannot turn thinking off on those four in Claude Code. Change it with /effort.

Use medium on a change you will review, and high when you want it to check its own work. Go higher only after medium misses.

Codex and Claude Code

Codex and Claude Code are the apps. The model inside is still Astra, Sol, Fable, Opus, or one of the cheaper ones.

  • Codex: the November 17 launch put GPT-6.2 Astra and Sol on Plus, Pro, Enterprise, and Edu. The Codex pricing page also lists Business. Enterprise and Edu need an admin to turn them on. CLI version 0.166.0 or later. For a normal day of coding, use Sol, and move to Astra when the change crosses several services.
  • Claude Code aliases: fable, opus, sonnet, haiku. opusplan uses Opus to plan and Sonnet to edit. Included on paid Claude plans. Chats and Claude Code share one usage pool.
  • ChatGPT Work and Codex usage is separate from ordinary Chat. Plus local-message estimates per five hours: Astra 6 to 50, Sol 17 to 180, Terra 35 to 285, Luna 350 to 3,000. OpenAI lists those as estimates, and weekly limits can still apply. Pro has no five-hour limit on that page.
  • Set permission rules and keep the branch, sandbox, and diff review. Moving an OpenClaw bot is a short config change. The model id in that walkthrough is older than this chart.
ChatGPT plans vs Claude plans

Plus and Claude Pro both cost $20. The models inside them differ.

  • Free. ChatGPT: GPT-5.6 Luna in Chat. Codex pricing lists GPT-6 Luna on desktop, still rolling out. Claude Free: Sonnet and Haiku. No Opus, no Fable, no Claude Code.
  • Go. Paid step under Plus. Chat still uses GPT-5.6 Luna. OpenAI’s January 2026 announcement listed $8 a month in the US, and the Go help article says the price can change. Closer to Claude Free than to Claude Pro.
  • Plus, $20. Chat: GPT-6 Sol at Medium and High. Work and Codex: GPT-6.2 Astra and Sol, with a small Astra allowance. Closest Claude plan is Pro: $20 a month, or $17 a month on the $200 annual plan. Opus, Sonnet, Haiku, and Claude Code. Fable only on usage credits.
  • Pro 100, $100. No five-hour Work and Codex cap. Chat adds Extra High and GPT-6 Pro (GPT-6 Astra). Closest Claude plan is Max, $100: 5x Pro usage, Fable for up to half the weekly limit, $100 a month in API credits.
  • Pro 200, $200. Larger allowance than Pro 100. Closest Claude plan is Ultra, $200: no five-hour limit for verified accounts, 10x weekly Pro usage, $200 in API credits. Fable still capped at half the weekly limit.
  • Pro 500, $500. Highest included usage. Ultrafast included. No individual Claude plan at this price on claude.com/pricing.

Sources: Plus, Pro tiers, GPT-6 in Chat, Codex pricing, the Astra and Sol announcement, and Claude pricing.

Ultrafast is the one disagreement. The November 17 announcement says Pro 500 can use it on Astra or Sol, and Pro 200 gets it starting with Sol. The Pro tiers article, written about GPT-6 Astra and GPT-6.1 Sol, says Ultrafast is Pro 500 only. If you are on Pro 200, check the picker.

If you already use Claude and want to try OpenAI, Plus is enough to see Sol and a little Astra. Going the other way, Claude Pro includes Opus. Fable on that plan runs on usage credits.

What OpenAI reported against Fable

On the November 17 launch post, OpenAI cites Artificial Analysis AA-3D at 1,318 Elo for Astra, 1,271 for Fable 5.2, and 1,259 for GPT-6.2 Sol. On Agents’ Last Exam they report Astra 5.5 points ahead of Fable, at about 5% of the cost, and they note that Fable’s leaderboard result left out nine tasks that used a fallback model. OpenAI published those numbers. I have not rerun the tests.

ChatGPT vs Claude models FAQ

What is the Claude equivalent of GPT-6.2 Astra?

Claude Fable 5.2, at the same $8 / $40. In ChatGPT, Astra is in Work and Codex. The chat button and the model name you use in the API (for example gpt-6.2-astra) can share a nickname and still be different models.

What is the Claude equivalent of GPT-6.2 Sol?

Claude Sonnet 5.5, also $2 / $10. Sonnet is on every Claude plan, including Free. GPT-6.2 Sol starts at Plus, in Work and Codex.

Sol vs Sonnet: which Sol?

The $2 / $10 pair is GPT-6.2 Sol and Sonnet 5.5. Pages that say Sol vs Sonnet often mean GPT-6 Sol, the earlier model behind the GPT-6 button in ordinary Chat. OpenAI's Chat help article says you pick the 6.2 models in Work or Codex.

What is the OpenAI equivalent of Claude Opus?

Nothing in the GPT-6.2 line is $4 / $20. GPT-5.6 Sol matches Opus 5.5 through November 30, 2026. For new work, start at GPT-6.2 Sol, which is half that token price.

What is the Claude equivalent of GPT-6 Luna?

Claude Haiku 5.5, at $0.10 / $0.50 until the prompt passes 100,000 tokens. After that Haiku is $0.50 / $2.50 for the whole request. Free and Go Chat currently run GPT-5.6 Luna.

Is there a Claude model like GPT-6.1 Terra?

No. Nothing on the current Claude ladder is $1 / $5. Terra is for screenshots, clicking a UI, and drafts, in Work and Codex on Plus and up. The regular Chat picker does not list it.

Which models do Codex and Claude Code use?

Codex runs GPT-6.2 Astra and Sol on Plus and above. The CLI needs version 0.166.0 or later. Claude Code aliases are fable, opus, sonnet, and haiku. opusplan uses Opus to plan and Sonnet to edit.

Do the chart prices cover a long agent prompt?

They cover the short-context rate. Past 272,000 input tokens, the OpenAI models here bill 2x input and 1.5x output for the whole request. Haiku moves to $0.50 / $2.50 past 100,000 prompt tokens.

When was this ChatGPT vs Claude map last updated?

November 20, 2026. GPT-6.2 Astra and Sol launched on November 17. GPT-5.6 Sol's $4 / $20 rate is promotional through November 30, 2026.

This page may contain affiliate links. Please see my affiliate disclaimer for more info.

Related Posts

View All Posts »