Gemini 3.5 Flash vs Kimi K3 for multi-step research
Kimi K3 wins, 88–78
Same 12 multi-step research tasks, marked blind by three rival labs. Kimi K3 took 10 tasks, Gemini 3.5 Flash took 1, 1 tied. Tested 18 Aug 2026.
Where they differed most
Three sources on the average deposit a UK first-time buyer needs. Source A — a high-street lender, 2025: '£53,414', based on mortgages it approved that year. Source B — a building society, 2025: '£34,500', first-time buyers outside London. Source C — a London estate agency, 2024: '£71,000'. These are usually written up as a contradiction. Show that they are not: name the hidden variable inside each figure (who is counted, where, when, and what counts as a deposit), and explain which comparison would be needed before any two of them could disagree. Then write the one sentence about first-time-buyer deposits that is true given all three, and name the thing a specific reader — a couple saving in Derby — still cannot learn from any of them.
These three figures are often presented as a contradiction, but they are actually measuring entirely different things. Here is the breakdown of the hidden variables that explain the variation, followed by what is needed for a true comparison, the single true sentence, and the missing information for the Derby couple. ### The Hidden Variables To understand why these figures coexist peacefully, we must look at the four hidden variables behind each number: #### Source A (£53,414) * **Who is counted:** Only first-time buyers who successfully a…
They look like a contradiction only if you assume they're three estimates of one number. They aren't — no pair of them shares a population, a geography, a statistic, and a definition, so there is no shared quantity for them to disagree *about*. Each figure has hidden variables baked in: **Source A — £53,414 (lender, 2025 approvals)** - **Who:** people *this one lender approved*. Declined applicants — who skew toward smaller deposits — are invisible, as are cash buyers and everyone still saving. It describes survivors of one credit process, not…
Task by task
| Task | Gemini 3.5 Flash | Kimi K3 |
|---|---|---|
| Four numbers, one high street | 8 | 9 |
| The trial that both worked and didn't | 8 | 9 |
| Break the brief into answerable steps | 8 | 9 |
| Corroboration that isn't | 8 | 9 |
| The disagreement is definitional | 6 | 8 |
| A forecast is not an outcome | 8 | 9 |
| The evidence table, as JSON | 7 | 8 |
| Funded, and still the best data | 7 | 9 |
| Write only the limits section | 9 | 8 |
| Not answerable from what you have | 9 | 9 |
| The statistic and the doorstep | 8 | 9 |
| Refuse the manufactured briefing | 8 | 9 |
Full receipts: Gemini 3.5 Flash, Kimi K3 · judges claude-sonnet-5, gpt-5.6-terra, grok-4.5
Questions people ask
Which is better for multi-step research: Gemini 3.5 Flash or Kimi K3?
Kimi K3 — it scored 88/100 against 78/100 on our 12-task multi-step research suite, winning 10 tasks to 1 with 1 tied. Every answer was marked blind by three judges from three rival AI labs.
How was this tested?
Both models answered the identical published multi-step research 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 multi-step research head-to-heads: Claude Sonnet 5 vs Kimi K3 · Claude Sonnet 5 vs Gemini 3.5 Flash · GPT-5.5 vs Kimi K3 · Gemini 3.5 Flash vs GPT-5.5 · GPT-5.6 Sol vs Kimi K3 · Gemini 3.5 Flash vs GPT-5.6 Sol
Full ranking: Best AI for multi-step research · model pages: Gemini 3.5 Flash, Kimi K3