Claude for Professionals (2026): Tested, Plans, Limits

What is Claude, and which models does it offer?

Claude is Anthropic’s AI assistant, available through an app and an API. Its current lineup includes models with different roles, plus a restricted model that is not a consumer option.

• Claude Fable 5.1: the flagship, aimed at demanding reasoning and long-horizon agentic work.
• Claude Opus 5.5: Anthropic’s recommended starting model.
• Claude Sonnet 5.5: the balanced tier.
• Claude Haiku 5.5: the fast tier.
• Claude Mythos 5.1: restricted to vetted cyber-defense and life-science organizations; not generally available.

Older versions remain visible on vendor pages. In particular, Claude Sonnet 5 appears in our tests below, but it is not the current Sonnet model.

Which Claude plan fits your work?

Start with Free for a trial, or consider Pro if you want access to Opus. For a firm, compare Team seat costs with Enterprise’s seat minimum and separate usage charges rather than assuming a business plan is automatically the right fit.

Prices checked 2026-10-19 on the official pricing page.

These are USD list prices for US visitors, excluding tax. Readers in the UK, Australia and India should check the vendor’s pricing page for their country because local prices, currency and tax differ.

Choose based on the model you need and work you can review, not the plan name alone. Before committing to annual billing, try representative tasks without confidential data. Our guides for accountants and lawyers provide profession-specific starting points.

How did Claude do in our tests?

Opus led the tested Claude models on criteria passed in each round, but that is a result for these prompts, not an overall ranking. We tested through the API command-line harness with tools, web and memory off, not through the consumer app; app results may differ.

The bench used 80 prompts across 10 professions, with 5 rubric criteria per prompt and 3 runs per model per prompt. Round 1 contained 30 English and 10 Spanish prompts. Round 2 contained 20 French and 20 Brazilian Portuguese prompts with deliberate traps, including a wrong total and a wrong unit conversion. We do not average the rounds.

Judge caveat: All tested Claude models and Gemini 4 Argon shared the judge gpt-6-astra, making their comparison like-for-like. The judges were not equally strict: on the same 30 sampled Claude runs in round 1, gpt-6-astra passed 93% of criteria and claude-opus-5-5 passed 98%. On the same 10 sampled runs in round 2, they passed 74% and 86%, respectively. Part of the Claude-versus-GPT gap on this site therefore comes from the grader, not the model.

Claude Fable 5.1 joined the bench on 2026-10-20 using the same locked prompts, inputs and rubrics. It did not score above Opus on these prompts: round 1 was lower and round 2 was within a point. Its flagship label is not evidence of better performance here.

Claude Sonnet 5.5, Claude Haiku 5.5 and Claude Mythos 5.1 are not in this test round yet.

Most lost points involved safety criteria: adding a warning, date, benefit or legal assumption absent from the input. Opus and the older Sonnet also lost points on English word limits.

In a separate Excel / Microsoft 365 task round of 10 new prompts (5 rubric criteria each, 3 runs per model, not pooled with rounds 1 and 2), Claude Opus 5.5 passed 93% of criteria with 22 of 30 runs passing every criterion, Claude Fable 5.1 passed 88% with 18 of 30, and Claude Sonnet 5 passed 83% with 14 of 30. On 6 sampled runs from this round, gpt-6-astra passed 93% of criteria and claude-opus-5-5 passed 97%. These were text-only answers: formulas were judged by their logic and the stated result on the given data, not executed in Excel, and the models were chat models answering a pasted task, not Copilot in Excel or Microsoft 365.

Spanish round 2 used a separate set of 10 new prompts for engineers, travel agents, construction managers, event planners and interior designers in Mexico, Spain and Colombia, not pooled with rounds 1 and 2 or the Excel round. Claude Opus 5.5 passed 83% of criteria with 11 of 30 runs passing every criterion, Claude Fable 5.1 passed 73% with 3 of 30, and Claude Sonnet 5 passed 72% with 8 of 30. These three models share the judge gpt-6-astra with Gemini 4 Argon, so their results are like-for-like; GPT’s are not, because it had a different judge. On 6 sampled runs from this round, gpt-6-astra passed 77% of criteria and claude-opus-5-5 passed 93%, and in a spot check an editor (the agent operating the bench, not a human) re-read 12 verdicts and agreed with 8.

On the Claude Skills bench, Opus and Sonnet averaged 4.0 to 5.0 of 5 criteria per skill, language and model. Fable used the same 20 skill-language pairs, with 3 runs each: 60 runs were graded, averaging 4.0 to 5.0 of 5, and 15 of 20 pairs averaged 5 of 5. In a separate French and Brazilian Portuguese Skills round, not pooled with the English and Spanish skills, Opus, Sonnet and Fable were tested on separately locked rubrics, with 4 adaptations revised after the first run and re-tested on the same inputs; current averages range from 4.33 to 5.0 of 5 per skill, language and model, with 46 of 60 combinations averaging 5 of 5 and no judge calibration or spot check. These results also used gpt-6-astra; the judge-asymmetry caution applies, and this Skills bench had no judge calibration or spot check. Our Claude Skills are published under CC BY 4.0, so they are free to use and adapt, including commercially, with attribution.

Read our testing methodology and test-round report before treating these results as a purchasing shortcut.

What should you check before pasting confidential data?

We have not verified the vendor’s data terms for this page. Read your exact plan’s data and privacy terms and your professional confidentiality rules before entering client, patient or employee data.

We make no claims about retention, training on your data or business-plan protections. A Team or Enterprise label is not enough to establish whether a plan meets your obligations. Keep that decision separate from whether you find its answers useful.

When should you not rely on Claude?

Do not use Claude as the final authority for figures, legal positions, clinical decisions or property facts. Check the underlying inputs and any assumptions it adds, rather than accepting a polished answer as evidence of accuracy.

Tax, legal, health and property work must be checked by a licensed professional in your jurisdiction. Use our prompt library for starting points, but still check facts, calculations and requested word limits yourself.

FAQ