Last verified

Kimi K3 vs Qwen3.8 Max — price, specs and shared benchmarks.

Kimi K3 lists $3.00/$15.00 per 1M input/output tokens; Qwen3.8 Max lists $2.00/$6.00. On our blended workload Kimi K3 works out 17% cheaper ($1.92 vs $2.32 per 1M). They share 7 benchmarks, all scored by each lab itself; Kimi K3 reports the higher number on 6.

Newer matchup

The current Kimi and Qwen flagships are Kimi K3.1 Pro vs Qwen4 Max. This page stays up for anyone still running Kimi K3 or Qwen3.8 Max; its prices are re-read with every snapshot.

Our read

Qwen3.8 Max is cheaper on both list rates, $2.00/$6.00 against Kimi K3's $3.00/$15.00. The blended row can still favour Kimi: Moonshot lists a cache-hit rate for K3 and Alibaba's card, as we last read it, lists none for this model, so our formula bills every Qwen input token at full price. Read that row as a gap in disclosure as much as in price.

  • Output-heavy work (long answers from short prompts) is where Qwen's lower output rate decides the bill.
  • Context that repeats across requests is where Kimi's cache rate pays off.
§ 01 / VERDICT

Who wins, row by row.

Category Winner Margin
Input price · per 1M tokens BQwen3.8 Max $2.00 vs $3.00 · 33% cheaper
Output price · per 1M tokens BQwen3.8 Max $6.00 vs $15.00 · 2.5× cheaper
Cache reads · per 1M cached tokens ·One on file Kimi K3 lists $0.30; no cache-read rate on file for Qwen3.8 Max
Effective cost · blended 92/8 · 82% cache AKimi K3 $1.92 vs $2.32 · 17% cheaper · Qwen3.8 Max billed at full input rate
Context window · max input tokens ·Even 1M on both
Benchmarks · shared published scores ·Self-reported 7 shared, all from each lab's own report; Kimi K3 is higher on 6
§ 02 / PRICING

What it actually costs.

Cost calculator →
Aspect Kimi K3 Qwen3.8 Max
API · inputper 1M tokens · from snapshot verified Oct 29 $3.00 $2.00 B wins
API · outputper 1M tokens · from snapshot verified Oct 29 $15.00 $6.00 B wins
API · cached inputper 1M cached tokens verified Oct 29 $0.30 —
API · batch inputper 1M tokens · batch verified Oct 29 $1.80 —
API · batch outputper 1M tokens · batch verified Oct 29 $9.00 —
Effective costblended 92/8 · 82% cache verified Oct 29 $1.92 A wins $2.32
Context windowmax input tokens even verified Oct 29 1M 1M
§ 03 / FEATURES

Specs, side by side.

Download CSV →
Capability Kimi K3 Qwen3.8 Max
Text ✓ Yes ✓ Yes
Image input ✓ Yes ✓ Yes
Video input ✓ Yes ✓ Yes
Code ✓ Yes ✓ Yes
Prompt caching ✓ $0.30 / 1M read ~ No rate on file
Batch API ✓ $1.80 in / $9.00 out ~ No rate on file
Released July 16, 2026 July 19, 2026
Status Active Active
Price checked November 18, 2026 October 29, 2026
§ 04 / BENCHMARKS

Shared benchmarks, same setting.

All benchmarks →
Coding · DeepSWE
Kimi K3
67.5%
Qwen3.8 Max
56.6%
each lab's own report · 27 ranked
Kimi K3
81.2%
Qwen3.8 Max
73.5%
each lab's own report · 5 ranked
Agents and computer use · Terminal-Bench 2 · Terminal-Bench 2.1
Kimi K3
88.3%
Qwen3.8 Max
86.6%
each lab's own report · 17 ranked
Agents and computer use · Toolathlon
Kimi K3
76.5%
Qwen3.8 Max
72.5%
each lab's own report · 9 ranked
Expert knowledge · GPQA Diamond
Kimi K3
93.5%
Qwen3.8 Max
92.6%
each lab's own report · 13 ranked
Kimi K3
56.0%
Qwen3.8 Max
56.2%
each lab's own report · 20 ranked
Professional work · JobBench
Kimi K3
54.3%
Qwen3.8 Max
53.4%
each lab's own report · 5 ranked

Frequently asked.

Common questions about this comparison, with sources where they matter.

Q · 01 Is Kimi K3 cheaper than Qwen3.8 Max? +
Kimi K3 costs $3.00 per 1M input tokens and $15.00 per 1M output tokens. Qwen3.8 Max costs $2.00 and $6.00. Qwen3.8 Max is cheaper on both. With our blended workload (92% input, 8% output, 82% of input from cache) the effective rates are $1.92 and $2.32 per 1M tokens. Both prices were read on the vendors' own pages; the older reading is from October 29, 2026.
Q · 02 Which has the bigger context window, Kimi K3 or Qwen3.8 Max? +
Neither: both accept 1M tokens of input.
Q · 03 Which scores higher on benchmarks, Kimi K3 or Qwen3.8 Max? +
DeepSWE: Kimi K3 67.5%, Qwen3.8 Max 56.6%; FrontierSWE · Version not stated: Kimi K3 81.2%, Qwen3.8 Max 73.5%; Terminal-Bench 2 · Terminal-Bench 2.1: Kimi K3 88.3%, Qwen3.8 Max 86.6%; Toolathlon: Kimi K3 76.5%, Qwen3.8 Max 72.5%; GPQA Diamond: Kimi K3 93.5%, Qwen3.8 Max 92.6%; Humanity's Last Exam · With tools: Kimi K3 56%, Qwen3.8 Max 56.2%; JobBench: Kimi K3 54.3%, Qwen3.8 Max 53.4%. All of these are the labs' own reports, not independent runs; each links to its leaderboard with the source and date.
Q · 04 Do Kimi K3 and Qwen3.8 Max support prompt caching? +
Kimi K3 has a cache-read rate of $0.30 per 1M tokens. We have no cache-read rate on file for Qwen3.8 Max, so our blended figure bills its input at the full rate.
Q · 05 When were Kimi K3 and Qwen3.8 Max released? +
Kimi K3 launched on July 16, 2026, Qwen3.8 Max on July 19, 2026.