Short answer: neither model wins across every workload. In Plotly’s single-run, 43-question data-analysis test, Qwen4-Max and Kimi K3 both scored 35/43. The reported run cost was $2.29 for Qwen4-Max and $5.01 for Kimi K3, but those totals reflect one specific agent, cache pattern and provider route. They are not a universal cost ratio.
Qwen4-Max became generally available on November 16, 2026. Moonshot’s current homepage also lists Kimi K3.1 Pro, announced later than the original Kimi K3. This comparison keeps the requested Kimi K3 version separate; check the K3.1 Pro page if you mean Moonshot’s current flagship.
Key takeaways
Plotly reported a tie at 35/43 in one run; its repeated runs on other models varied by up to three points, so the tie does not prove identical performance.
Qwen4-Max is newer: Alibaba announced availability on November 16, 2026.
Both vendors publish a 1M-token context window, but that alone says nothing about accuracy, latency or the price of a long request.
Moonshot’s homepage lists Kimi K3.1 Pro, dated October 24, 2026. Do not treat Kimi K3 benchmark results as results for K3.1 Pro.
Choose with repeated tests of your own prompts, tools, regions and cache conditions.
Qwen4-Max and Kimi K3 at a glance
| Published detail | Qwen4-Max | Kimi K3 |
|---|---|---|
| Provider model ID | qwen4-max | kimi-k3 |
| Context published by provider | 1M tokens | 1M tokens |
| Plotly score, one run | 35/43 | 35/43 |
| Plotly run cost | $2.29 | $5.01 |
| Median time per question | 40 seconds | 43 seconds |
The Plotly figures come from a 43-question data-analysis benchmark run once per model in Plotly Studio Embedded. The question set and dataset are private. Plotly says repeated tests of other models varied by up to three points. Treat the equal scores as a narrow tie, not proof that the models perform identically. The run cost includes its task, tools, provider route and cache use; it is not a general per-request saving.
What the published prices do—and do not—tell you
Alibaba’s Singapore international API pricing, checked November 17, 2026, lists Qwen4-Max at $1.50 input and $4.50 output per million tokens. Moonshot’s Kimi K3 announcement, dated July 22, lists $3 uncached input, $0.30 cached input and $15 output per million tokens. These are provider rate cards from different dates and routes. They do not include a service subscription, retries, tools, or your actual cached-token share.
| Cost question | Public evidence | Limit |
|---|---|---|
| Which cost less in Plotly’s run? | Qwen4-Max: $2.29; Kimi K3: $5.01 | Not a general cost ratio for other workloads |
| Which has lower listed input and output rates? | Qwen4-Max in the cited Singapore rate row | Cache, region, output volume and discounts change totals |
| What is Kimi K3’s cached input rate? | Moonshot lists $0.30 per million tokens | Gateway rates and cache ratios can differ |
For a fair procurement test, fix model IDs, system prompts, tools, region and cache policy. Run the same representative tasks several times. Record correctness, retries, human corrections, completion time, input/output/cache tokens and the final invoice. A benchmark cost per run is useful only when the task, harness and accounting method match your own use.
Which model should you try first?
If your work resembles Plotly’s private data-analysis exam, the public result shows equal single-run scores and a lower Qwen4-Max run total. It does not establish a stable quality or cost advantage. The benchmark had 43 questions, a tool-using agent and one run per model.
For long-context research, both vendors state a 1M-token window. Test retrieval and factual coverage on your own documents; a context limit does not guarantee every detail will be attended to. For coding, agentic work or multimodal tasks, use a task-specific benchmark instead of extrapolating from the Plotly analytics test.
The exact model version matters. Alibaba announced Qwen4-Max on November 16, 2026. Moonshot’s current model list lists Kimi K3.1 Pro, dated October 24, ahead of the original Kimi K3 from July. If you are selecting Moonshot’s current flagship, compare K3.1 Pro separately. Do not use the older K3 result as if it measured K3.1 Pro.
A practical comparison workflow
Use the same prompt and source material when you test a model on your own work.
For model-specific next steps, see the Qwen4-Max hub, Kimi K3 hub, agentic browser guide, and Kimi K3 pricing. For evaluation methodology, read our agent cost-per-success guide, source-backed browser research guide, AI research workflow guide, and AI browser comparison.
Verdict
The strongest public head-to-head result found is a one-run tie on Plotly’s data-analysis benchmark: both models scored 35/43. Qwen4-Max had the lower reported run cost and slightly lower median time, but the private data, one-run design and documented variance rule out a universal winner claim. Start with the model whose endpoint, regional pricing and access terms fit your deployment, then test real prompts several times. If Moonshot is on your shortlist, check whether Kimi K3.1 Pro—not Kimi K3—is the version you intend to evaluate.
Sources
Tabbit Browser offers Qwen4-Max. Download Tabbit, then select Qwen4-Max from the model list in Chat to start using it.
FAQ
Which model scored higher on Plotly’s benchmark?
Neither. Both scored 35 of 43 questions in one run; Plotly cautions that single-run scores can vary.
Which model is cheaper?
In Plotly’s reported run, Qwen4-Max cost $2.29 and Kimi K3 cost $5.01. These are run totals, not general price guarantees.
Do both models support a one-million-token context?
Both vendors publish a 1M-token context figure. Check the exact endpoint’s current limits and billing.
Is Kimi K3 Moonshot’s newest model?
Moonshot also lists Kimi K3.1 Pro, dated October 24, 2026. This article keeps Kimi K3 as a separate version.
How should I choose between them?
Repeat representative tasks with fixed prompts, tools, model IDs, region and cache setup; compare accuracy, completion and total bill.