model comparison

vs

Qwen3.8 Max vs Kimi K3

Qwen against Moonshot AI, compared on listed price, real workload cost, context and capabilities.

the short answer

  • on Terminal Bench 2.1, as each maker reports it, Qwen3.8 Max scores 86.6 and Kimi K3 88.3.
  • Qwen3.8 Max lists lower input and output prices.
  • Qwen3.8 Max reads cached input for less ($0.25/M against Kimi K3's $0.3/M), which decides the bill for agents that resend long context.

side-by-side specs

specQwen3.8 MaxKimi K3
providerQwenMoonshot AI
input price$2/M$3/M
output price$6/M$15/M
cached input$0.25/M$0.3/M
blended price (3:1)$3/M$6/M
context window1M tokens1M tokens
max output131k tokens—
inputtext, images, videotext, images
structured outputyesyes
weightsopen (qwen3.8-max)open (Kimi K3 License)
releasedAugust 2026July 2026
Terminal Bench 2.1 (maker-reported)86.688.3
GPQA Diamond (maker-reported)92.693.5

same work, both models

workloadQwen3.8 MaxKimi K3
one chat reply3,000 tokens in, 500 outKimi K3 costs 1.8× as much as Qwen3.8 Max$0.009$0.02
summarise a 100-page report50,000 tokens in, 1,500 out, one requestKimi K3 costs 1.6× as much as Qwen3.8 Max$0.11$0.17
1,000 classification calls500 tokens in, 20 out, per call$1.12$1.80
coding-agent session2M tokens in across ~60 turns, 60,000 out, no cache$4.36$6.90
same session, cached85% of input read from the prompt cache, the rest written to itKimi K3 costs 1.7× as much as Qwen3.8 Max$1.39$2.31

at listed rates. token counts are typical sizes, not measurements.

Qwen3.8 Max pricing notes

  • context caching is implicit: cache hits bill at $0.25 per 1M, cache writes at the normal input rate.

Kimi K3 pricing notes

  • not included in any plan: every request bills from credits.
  • context caching is automatic: cache hits bill at $0.30 per 1M input tokens, a tenth of the input rate.
  • thinking is on by default at max effort, and reasoning tokens bill at the output rate.

when to pick which

Qwen3.8 Max

Qwen3.8 Max is Alibaba's flagship Qwen3.8 model: a hosted 2.4T-parameter MoE that takes text, image and video, with a thinking mode. it succeeds Qwen3.7 Max at lower rates and adds image input; its open weights are text-only.

pick it if

  • you want Qwen3.8's flagship with image and video input
  • long agent sessions reuse a prefix; cache hits bill $0.25 per 1M
  • you want a Max-class model whose text-only base weights are downloadable

skip it if

  • Qwen4 Max fits the task; same window and inputs, and it bills less on input, output and cache hits
  • Qwen3.8 Omni Flash covers the task; it takes images, video and audio, far cheaper

Kimi K3

Moonshot AI's most capable Kimi model: an open-weights 2.8T-parameter mixture-of-experts with native vision and a 1M-token context. it sits above Kimi 2.6 and Kimi 2.7 Code, which have a 256K context and cost less per token.

pick it if

  • you want the model Moonshot calls its most capable
  • your prompts outgrow the 256K context of Kimi 2.6 and 2.7 Code
  • you resend long prompts, so cache hits bill at a tenth of input

skip it if

  • Kimi 2.6 or 2.7 Code covers your workload at a lower rate

faq

which is better, Qwen3.8 Max or Kimi K3?

on Terminal Bench 2.1, as each maker reports it, Qwen3.8 Max scores 86.6 and Kimi K3 88.3. these are each maker's own numbers, not a controlled head-to-head; the spec table lists every benchmark both publish.

which is cheaper, Qwen3.8 Max or Kimi K3?

Qwen3.8 Max lists lower input and output prices. on a coding-agent session (2M tokens in, 60,000 out) Qwen3.8 Max costs $4.36 and Kimi K3 $6.90.

can i switch between Qwen3.8 Max and Kimi K3 mid-task?

yes. in an octomind session, /model octohub:qwen3.8-max and /model octohub:kimi-k3 switch models without losing the conversation, so you can draft on one and review on the other.

try either model

switch between models inside the same octomind session.

prices from the catalog of 2026-11-18