GLM 5.2 vs Kimi K3 for workflow automation
Kimi K3 wins, 85–82
Same 12 workflow automation tasks, marked blind by three rival labs. Kimi K3 took 4 tasks, GLM 5.2 took 0, 8 tied. Tested 18 Aug 2026.
Where they differed most
A 40-person UK company wants to automate expense reimbursement end to end: receipt photographed, read, categorised, approved, paid into the employee's bank account, posted to the ledger. The finance director's brief is 'no human touches it under £100'. Go through the process and mark, for each stage, whether it can be automated, automated with a check, or must stay with a person — and give the reason in terms of what goes wrong, not in terms of policy. Name the two stages where full automation would be genuinely dangerous for a company this size, and say what the cheapest sufficient control is at each: not the most thorough control, the cheapest one that actually catches the thing.
Here is the breakdown of the end-to-end expense process, evaluated by failure modes rather than policy, followed by the two genuinely dangerous stages and their cheapest sufficient controls. ### Process Breakdown **1. Receipt photographed** * **Status:** Automated * **What goes wrong:** The photo is blurry, poorly lit, or cuts off the total. * **Reason:** Modern smartphone camera APIs can enforce edge detection, glare reduction, and auto-focus. If the image doesn't pass these basic quality thresholds, the app simply refuses to accept t…
## Stage by stage **1. Capture (photo → submission) — Automate, with one structural rule.** What goes wrong: the same receipt submitted twice; two colleagues at the same meal each claiming the whole bill; and fabricated images — a generated receipt is now a five-minute job. Duplicate detection (perceptual hashing plus amount+date+vendor matching across *all* employees) is fully automatable and catches the first two. For fakes, the fix is in how you build capture, not in checking: force live in-app camera photos, no gallery uploads. Faking then…
Task by task
| Task | GLM 5.2 | Kimi K3 |
|---|---|---|
| The MOT reminder nobody sends | 8 | 8 |
| The spec, as JSON | 8 | 8 |
| Where the human stays | 8 | 9 |
| The unhappy paths are the job | 8 | 8 |
| Do not automate this | 9 | 9 |
| It fired twice | 8 | 9 |
| The spreadsheet that runs the business | 8 | 8 |
| The alert that wakes someone up | 8 | 8 |
| How long it really takes | 8 | 9 |
| No API, no chance? | 8 | 8 |
| Rules, not guesses | 8 | 9 |
| Refuse the scraper | 9 | 9 |
Full receipts: GLM 5.2, Kimi K3 · judges claude-sonnet-5, gemini-3.1-pro-preview, gpt-5.6-terra
Questions people ask
Which is better for workflow automation: GLM 5.2 or Kimi K3?
Kimi K3 — it scored 85/100 against 82/100 on our 12-task workflow automation suite, winning 4 tasks to 0 with 8 tied. Every answer was marked blind by three judges from three rival AI labs.
How was this tested?
Both models answered the identical published workflow automation tasks. Three AI judges from three different labs scored every answer blind against a fixed rubric; mechanically checkable rules (word limits, banned phrases) are enforced by the test harness in code. The raw outputs and judge verdicts are downloadable.
More workflow automation head-to-heads: GPT-5.6 Sol vs Kimi K3 · GLM 5.2 vs GPT-5.6 Sol · Claude Sonnet 5 vs Kimi K3 · Claude Sonnet 5 vs GLM 5.2 · GPT-5.5 vs Kimi K3 · GLM 5.2 vs GPT-5.5
Full ranking: Best AI for workflow automation · model pages: GLM 5.2, Kimi K3