Opus costs exactly twice what Sonnet does per token: $4 against $2 per million in, $20 against $10 out. Per finished task the gap is far smaller, because Opus uses fewer tokens, and one independent run at max effort put them about 10% apart. Sonnet is faster and on the Free plan. Opus is what Claude Code picks for you.
also searched as: claude opus vs sonnet · opus vs sonnet · sonnet vs opus · claude sonnet vs opus · sonnet vs opus claude · opus vs sonnet claude · opus or sonnet
Pick Claude Opus if the work is long, open-ended or spread across many files, because that is where Anthropic says Opus stays clearly ahead, and at high effort it costs about what Sonnet does per task.
Pick Claude Sonnet if you are on the Free plan, want faster answers, or run scoped fixes and documents at low or medium effort, where Sonnet costs less per task.
Pick neither if the job is summaries, classification or subagent grunt work, where Haiku costs a twentieth of Sonnet per token on short prompts, or a multi-hour agent run that Opus at max effort still fails, which is Fable's job and bills to usage credits on Pro.
prices checked 2026-10-13 · by dan scutaru
| Claude Opus | Claude Sonnet | |
|---|---|---|
| price | from $20/mo | free tier · from $20/mo |
| free tier | None. Claude's Free plan offers Sonnet and Haiku, not Opus. | Sonnet is on Claude's Free plan alongside Haiku, for chat on the web and mobile apps, on a rolling five-hour usage window with no published message count. Claude Code is not included on Free. |
| runs on | web, ios, android, mac, windows, linux, api, amazon bedrock, google cloud, microsoft foundry, github copilot | web, ios, android, mac, windows, linux, api, amazon bedrock, google cloud, microsoft foundry, github copilot |
| API price per million tokens | $4 in, $20 out (Opus 5.5). Half that on the Batch API. Fast mode $8 / $40 | $2 in, $10 out (Sonnet 5.5). Half that on the Batch API. No fast mode |
| context window | 1M tokens in the API, Claude Code and paid apps. 128K max output | 1M tokens in the API, Claude Code and paid apps. 128K max output |
| speed | Anthropic rates it Moderate. Fast mode is up to 2.5x quicker, on usage credits only for Pro and Max | Anthropic rates it Fast. It publishes no tokens-per-second figure for either |
| Claude plans that include it | Pro, Max, Team, Enterprise. Not Free | Every plan, Free included |
| how fast it uses your Pro or Max limit | No published rate. Anthropic says the model you pick changes how far the limit goes. Past it, usage credits charge $4 / $20 | No published rate either. Past the limit, usage credits charge $2 / $10 |
| default in Claude Code | Yes, on Pro, Max, Team, Enterprise, the API, Bedrock and Google Cloud | Only on Microsoft Foundry, where it is Sonnet 4.5. Elsewhere you switch with /model sonnet |
On Pro, Max, Team, Enterprise and the API, Claude Code starts on Opus 5.5 by default. Anthropic's model guide agrees and says to start with Opus for most workloads. Its Claude Code cost page says the opposite: Sonnet handles most coding tasks well and costs less.
The launch numbers explain the split. Sonnet 5.5 beat Opus 5.5 on Terminal-Bench 4.0, 70.6% to 66.4%, and trailed on CursorBench, 55.5% to 57.8%. Give Sonnet scoped fixes and tests, and Opus refactors across many files. On Bedrock or Google Cloud, Claude Code's sonnet alias still means Sonnet 4.5 unless you pin the model ID.
Per token it is a flat 2x: Opus 5.5 is $4 in and $20 out per million tokens, Sonnet 5.5 is $2 and $10, and GitHub Copilot bills the same rates. Per task it is far closer. Artificial Analysis ran both at max effort: Opus used 260M output tokens to Sonnet's 420M and cost $5.98 per task against $5.46.
Effort moves the bill more than the model. Anthropic says Sonnet saves money at low or medium effort and matches Opus at a similar cost at high. On the API Sonnet defaults to high effort and Opus to medium, so compare them at the same setting.
Anthropic publishes no ratio. It says the model, the effort level and the length of the chat decide how far your limit goes, and the five-hour and weekly limits count every model together. Chat and Claude Code share that pool, as our Claude vs Claude Code page explains. Claude Code can also show a separate Opus or Sonnet limit; hit one and you can carry on with the other.
Past the limit, usage credits bill at API rates, so there Opus costs twice as much. If you hit the limit most days, our Claude Pro page weighs Pro against Max.
Haiku 5.5, released on 7 October 2026, costs $0.10 in and $0.50 out per million tokens for prompts up to 100K tokens, a twentieth of Sonnet. Anthropic pitches it for summaries, classification and subagent work, and says Sonnet and Opus remain better for complex agentic coding. It is on the Free plan too.
Above Opus sits Fable 5.1, at $10 in and $50 out. On Pro it runs on usage credits from the first message, and Max can spend up to half its weekly limit on it. Anthropic says Opus 5.5 matches Fable 5.1 on most work, so try Opus at max effort first.
On hard, open-ended work, yes, and Anthropic says that is where the gap is clearest. On scoped tasks the launch numbers are close, and Sonnet won one agentic coding test outright. Try the same task on both at the same effort first.
Opus 5.5 for most people, which is why Claude Code defaults to it on every paid plan. Sonnet 5.5 when speed matters or the tasks are small. Fable 5.1 only for long agent runs Opus cannot finish. Either way, Claude Code needs a paid plan or an API key.
No. The Free plan has Sonnet and Haiku only. Opus starts with Pro at $20 a month, or per token through the API. Our Claude Free page covers what the free plan does include.
Yes. Anthropic labels Sonnet 5.5 Fast and Opus 5.5 Moderate, without a tokens-per-second figure. Opus has a fast mode, up to 2.5x quicker at twice its normal price, billed only to usage credits on Pro and Max.