Best AI for HR and employment questions
Scored 94/100 on our 12-task HR suite — level on quality with Claude Sonnet 5 (94) — the top spot goes on the tie-break: cleanest rule-compliance, then lowest measured cost; GPT-5.5 sits a single point behind.
GPT-5.6 Luna is the engine inside ChatGPT. Go to chatgpt.com ↗ — the free tier is fine to start. Paid plans start at £7/month (go plan, vendor’s own price). The free tier is ChatGPT’s, not a promise about this exact model — we haven’t verified which plan carries it. Not fussed about the last point or two? Any of the top 3 here will serve you well.
Every task quotes the relevant public guidance in the prompt and is marked against that excerpt, because the failure mode here is a confident statutory figure that is out of date. The scenarios are the ones that actually reach an HR inbox: 'she's still on probation, we can just let her go', a manager who wants to ring a signed-off employee daily and ask what the doctor said, a grievance about the manager who would normally hear it, and a flexible-working refusal whose reason is 'everyone will want it'. The sharpest task hands the model a real-sounding percentage and asks it to confirm the figure so corrections can be run tonight — the right answer refuses to supply a number the guidance does not give and says where to check. None of this is employment-law advice — answers are judged against the guidance quoted in the task, and a good one tells you when to take proper advice. One task asks for a redundancy engineered around a pregnancy announcement, and the right answer is to refuse and name it.
updated 27 Aug 2026 · tested by Robert Prime · re-ranks automatically when a new run lands
- All 18 sat the identical 12-task suite — same tasks, same order, one attempt each.
- Every answer marked blind: judges are not told which entrant wrote it.
- Three judges per answer, each from a competing lab. A panel never includes the entrant's own lab.
- Scored 0–10 against a fixed rubric. Answers breaking a task's explicit rules are capped by machine, not by opinion.
- Ties broken by fewest rule breaches, then lowest measured cost — published, not editorial.
- Nobody pays for placement. Last computed 27 Aug 2026.
Every score below links to the raw file behind it: every task, the entrant’s real answer, and all three judges’ marks.
| # | Model | Our score |
|---|---|---|
| 1 | GPT-5.6 Lunalatest | 94/100 |
| 2 | Claude Sonnet 5latest | 94/100 |
| 3 | GPT-5.5 | 93/100 |
| 4 | GPT-5.3-Codex | 92/100 |
| 5 | GPT-5.6 Terra | 91/100 |
| 6 | GPT-5.6 Sol | 91/100 |
| 7 | Claude Opus 4.8 | 91/100 |
| 8 | DeepSeek V4 Pro | 90/100 |
| 9 | Kimi K3 | 90/100 |
| 10 | Grok 4.5 | 88/100 |
| 11 | Qwen3.7 Max | 88/100 |
| 12 | Gemini 3.1 Pro Preview | 87/100 |
| 13 | Claude Opus 4.6 | 86/100 |
| 14 | Gemini 3.1 Flash Lite | 85/100 |
| 15 | Gemini 3.5 Flash | 84/100 |
| 16 | GLM 5.2 | 83/100 |
| 17 | DeepSeek V4 Flash | 82/100 |
| 18 | Mistral Medium 3.5 | 78/100 |
Claude Fable 5 refused to attempt this suite, so it has no score to rank. Its provider returned a refusal rather than an answer — model refused (finish_reason=content_filter, native_finish_reason=refusal) — on 27 August 2026. This is the model’s own answer, not a test we skipped.
“API cost” is what software developers pay to build on a model — ignore it if you just use the website. Each model answers each task once. Models level on score are ranked by a fixed tie-break — fewest machine-checked rule breaches, then lowest measured cost per run — so the order is deterministic and checkable, never arbitrary. Judge panels never include the contestant’s own lab, so panels differ slightly per model — small cross-model gaps can reflect panel severity, not quality.
Made by OpenAI — their newest model. You use it inside ChatGPT — nothing to install.
Strongest showing: “The grievance about the manager” — scored 10/10 by the panel. Weakest: “Write only the invitation letter” at 9/10.
“The response perfectly follows all instructions, including the strict formatting constraints. It accurately extracts the 5 steps from the excerpt and logically addresses the complication with Dean as requested, without adding unnecessary in”— google/gemini-3.1-pro-preview, judging blind · full receipts ↓
12 tasks · 27 Aug 2026 · judges claude-sonnet-5, gemini-3.1-pro-preview, grok-4.5 · API $0.2 in / $1.2 out per 1M tokens · full model page →
Made by Anthropic — their newest model. You use it inside Claude — nothing to install.
Strongest showing: “The grievance about the manager” — scored 10/10 by the panel. Weakest: “The decision was made in the car” at 9/10. On 4 tasks a judge flagged a possible instruction breach — advisory only: the machine check did not confirm it and the panel outvoted it.
“The response perfectly follows all instructions, including formatting constraints. It accurately extracts the 5 steps from the excerpt and logically addresses the conflict of interest regarding Dean as requested, without any preamble.”— google/gemini-3.1-pro-preview, judging blind · full receipts ↓
12 tasks · 18 Aug 2026 · judges gemini-3.1-pro-preview, gpt-5.6-terra, grok-4.5 · API $2 in / $10 out per 1M tokens · full model page →
Made by OpenAI. You use it inside ChatGPT — nothing to install.
Strongest showing: “The grievance about the manager” — scored 9/10 by the panel.
“The response follows all constraints perfectly, providing exactly 5 bullets with no preamble. It accurately adapts the excerpt to the specific scenario involving Priya and Dean, logically addressing the conflict of interest.”— google/gemini-3.1-pro-preview, judging blind · full receipts ↓
12 tasks · 18 Aug 2026 · judges claude-sonnet-5, gemini-3.1-pro-preview, grok-4.5 · API $5 in / $30 out per 1M tokens · full model page →
See it for yourself
One task from the suite, and what the leader, the middle of the table and the bottom actually wrote. This is the task they disagreed on most — scores ranged from 4 to 10 out of 10.
Guidance excerpt: 'A fair grievance procedure normally follows these stages: the employee sets out the grievance in writing; the employer holds a meeting to discuss it without unreasonable delay; the employee has a statutory right to be accompanied at that meeting by a colleague or a trade union representative; the employer decides what action to take and confirms the decision to the employee in writing; and the employee is told they may appeal, with the appeal heard by someone not previously involved wherever possible. Written records should be kept at every stage.' Priya has submitted a grievance about her own line manager, Dean — who is the person who would normally hold the meeting. Using only the excerpt, set out the first 5 steps in order: exactly 5 bullets, one line each, no sub-points, no preamble. On the line where it matters, say what Dean's involvement changes and who should act instead.
- Priya sets out her grievance in writing. - The employer holds a meeting without unreasonable delay; because the grievance concerns Dean, someone else should hold it instead. - Priya may be accompanied at the meeting by a colleague or trade union representative. - The employer decides what action t…
Judge: The response perfectly follows all instructions, including the strict formatting constraints. It accurately extracts the 5 steps from the excerpt and logically addresses the complication with Dean as requested, without adding unnecessary in
- The employee sets out the grievance in writing - The employer holds a meeting to discuss it without unreasonable delay (Dean's involvement means he cannot hold it so another manager must act instead) - The employee has a statutory right to be accompanied at that meeting by a colleague or a trade u…
Judge: Accurate, follows 5-bullet format, correctly addresses Dean's conflict; excerpt doesn't explicitly state substitute manager but reasonable inference. Minor: no explicit basis for substitution in text.
- The employee sets out the grievance in writing. - The employer holds a meeting to discuss it without unreasonable delay. - The meeting is held by someone other than Dean, as he is the subject of the grievance. - The employee has a statutory right to be accompanied at that meeting by a colleague or…
Judge: Task requires exactly 5 bullets mapping the 5 stages with the Dean note embedded on the relevant line, not as an extra bullet. Response has 5 bullets but only 4 correspond to the excerpt's stages; missing the appeal step and adds an extra b
Trimmed to fit. Every full answer, every judge’s score: GPT-5.6 Luna, Grok 4.5, Mistral Medium 3.5
How this ranking is made
Every model answers the identical published task suite. Three AI judges from three different labs mark each answer blind against a fixed 0–10 rubric — none is ever from the maker’s own lab, and an answer that breaks a task’s explicit rules is capped in code, not by the judge’s goodwill. The score is the average, out of 100.
What this is not: an opinion piece, a paid ranking, or a benchmark we cannot show you. No vendor can buy inclusion, a position or a score on this page — the order is computed from the test results before any link to a product exists, and where a link earns us a commission it says so on the link itself and the order is identical either way. Every score links its raw outputs and judge verdicts. The full protocol · How we make money · receipts: GPT-5.6 Luna, Claude Sonnet 5, GPT-5.5, GPT-5.3-Codex, GPT-5.6 Terra, GPT-5.6 Sol, Claude Opus 4.8, DeepSeek V4 Pro, Kimi K3, Grok 4.5, Qwen3.7 Max, Gemini 3.1 Pro Preview, Claude Opus 4.6, Gemini 3.1 Flash Lite, Gemini 3.5 Flash, GLM 5.2, DeepSeek V4 Flash, Mistral Medium 3.5
Questions people ask
What is the best AI for HR and employment questions in 2026?
GPT-5.6 Luna leads our tested ranking with 94/100 on our 12-task HR suite, in a dead heat with Claude Sonnet 5 (94). Every answer was marked blind by three AI judges from three different labs, and the full outputs are downloadable.
How is this ranking made?
Each model answers the identical published task suite; three judges from different labs score every answer 0–10 against a fixed rubric without knowing which produced it; answers that break a task's explicit rules are capped automatically. The score is the average, out of 100. No vendor pays for placement.
What happens when two models score the same?
They are separated by a fixed, published tie-break rather than by editorial choice: first the fewest machine-checked rule breaches, then the lowest measured cost per run. The order is therefore reproducible from the same data, and where the tie-break decides the top place the page says so.
Did any model refuse to take this test?
Yes — Claude Fable 5 declined this suite outright, returning a refusal rather than an answer, so it has no score to rank. That is the model's own answer and not a test we skipped; the provider's refusal reason and the date are printed on the page.
How often does this page update?
It re-ranks itself whenever a new test run lands, and prices re-verify daily against vendor pages. The current ranking was last computed on 27 Aug 2026.
Head-to-heads in hr & employment
Show all 20 tested pairs ▾
More rankings ▾
Best AI for writing · Best AI chatbot for everyday use · Best AI for coding · Best free AI model · Best AI for spreadsheets and Excel · Best AI essay writer · Best AI for summarising documents · Best AI for extracting data from text · Best AI for writing emails · Best AI for everyday maths and percentages · Best AI for customer service replies · Best AI for revision and study notes · Best AI for vibe coding · Best AI for making flashcards · Best AI for social media posts · Best AI for job applications and cover letters · Best AI for presentations · Best AI for writing your CV · Best AI for research skills · Best AI for creative writing · Best AI for translation · Best AI for travel planning · Best AI for emotional support · Best AI for everyday legal questions · Best AI for health questions · Best AI for writing a book · Best AI for humanising AI text · Best AI research agent · Best AI for code review · Best AI for workflow automation · Best AI for bookkeeping · Best AI for meeting notes · Best AI for property and lettings · Best-value AI model API · every model we track · every tool