// head_to_head

Kimi K2 Thinking vs Qwen3.8 27B

Published LiveBench scores across all seven categories, live list pricing, context windows, and the measured cost of a point of capability — for both models, side by side.

OpenRouter + LiveBenchAll comparisonsFull leaderboard

Kimi K2 Thinking and Qwen3.8 27B are priced within ~10% of each other.

Kimi K2 Thinking does not have a published LiveBench run, so this comparison covers price, context and declared capabilities only. A missing score means "not evaluated", not "worse" — the right way to separate these two is an eval on your own workload.

moonshotai

Kimi K2 Thinking

Blended / 1M
$1.07
Context
262K
Released
Nov 6, 2025
Overall score
Not evaluated
reasoningtool callingprompt caching

qwen

Qwen3.8 27B

Blended / 1M
$1.06
Context
1M
Released
Aug 14, 2026
Overall score
75.3
reasoningtool callingimage inputvideo inputprompt caching

Specs and pricing

MetricKimi K2 ThinkingQwen3.8 27B
LiveBench overall

Mean of the seven LiveBench category scores, 0–100. Higher is better.

75.3
Cost per point

Measured benchmark spend divided by overall score — dollars per point of capability.

$0.0556
Blended price / 1M

3:1 input:output mix, the usual shape of production traffic.

$1.07$1.06
Input price / 1M$0.600$0.420win
Output price / 1M$2.50win$3.00
Cached input / 1M

Price of an input token served from the prompt cache, where the provider publishes one.

$0.150$0.085win
Context window262K1Mwin
Max output tokens98K131Kwin

Benchmarks by category

An overall score averages away the thing you probably care about. These are the seven LiveBench categories scored separately — Kimi K2 Thinking on top, Qwen3.8 27B below, both out of 100.

Agentic coding
61.4
Coding
75.7
Reasoning
80.0
Mathematics
86.2
Data analysis
76.6
Language
74.3
Instruction following
72.7

What each one costs to run

Per-token prices are hard to feel. These are monthly list costs for both models across five workload shapes, using each provider's published cached-input rate where there is one.

WorkloadKimi K2 ThinkingQwen3.8 27B
Support chatbot

1.2K in / 400 out × 200K requests

$311.60/mo$316.68/mo
RAG assistant

8K in / 600 out × 100K requests

$450.00/mo$382.00/mo
Coding agent

40K in / 4K out × 20K requests

$428.00/mo$388.40/mo
Document extraction

20K in / 1.5K out × 50K requests

$765.00/mo$628.25/mo
Bulk classification

500 in / 20 out × 5M requests

$1,525/mo$1,182/mo
Run these two through the cost calculator

Which should you pick?

You need to fit large documents in one call

Qwen3.8 27B

Wider context window — 1M against 262K.

Kimi K2 Thinking vs Qwen3.8 27B FAQ

Which is better, Kimi K2 Thinking or Qwen3.8 27B?

Kimi K2 Thinking and Qwen3.8 27B are priced within ~10% of each other. Kimi K2 Thinking does not have a published LiveBench run, so this comparison covers price, context and declared capabilities only. A missing score means "not evaluated", not "worse" — the right way to separate these two is an eval on your own workload.

Is Kimi K2 Thinking cheaper than Qwen3.8 27B?

They cost about the same. Both land near $1.07 per million tokens on a 3:1 input:output blend, so price is unlikely to be the deciding factor between them.

Does Kimi K2 Thinking or Qwen3.8 27B have a bigger context window?

Qwen3.8 27B has the larger context window: 262K for Kimi K2 Thinking against 1M for Qwen3.8 27B. Note that a window you can fill is not a window you should fill — retrieval quality usually degrades well before the limit, and you pay for every token you put in it.

Do Kimi K2 Thinking and Qwen3.8 27B support prompt caching?

Both publish a cached-input rate: $0.150 per million for Kimi K2 Thinking and $0.085 for Qwen3.8 27B, against full input rates of $0.600 and $0.420. On a workload with a long stable prefix — a system prompt, a tool schema, a retrieved corpus — that changes the economics more than the headline price does.

Related comparisons

How these numbers are produced

  • Price — provider list price from OpenRouter, refreshed every 15 minutes. “Blended” is a 3:1 input:output mix.
  • ScoresLiveBench release 2026-06-25, using their own category map. Each model shows its strongest published run. A blank means “not evaluated”, never “bad”.
  • Cost per point — the measured dollars LiveBench spent on the run, divided by the score it earned.
  • “Win” — awarded only past a threshold: one full point on a benchmark score, 10% on a price, 25% on a context window. Anything tighter reports as a tie, because effort settings alone move a LiveBench score by more than that.

Published benchmarks rank models on someone else's tasks. Before committing, see LLM & agent evaluation for building an eval on your own.