Kimi K2.5 vs Qwen2.5 32B Instruct.
model vs model · prices as of
Kimi K2.5 takes 3 of 5 rounds against Qwen2.5 32B Instruct. Qwen2.5 32B Instruct counters on price: $1.72 per million tokens against $2.80, for 131k context instead of 262k.
Kimi K2.5
$2.80/M tokens
cheapest we can explain, at NanoGPT · $0.294 per 1,000 pages
262k context · tool calling · reads images · open weights
Qwen2.5 32B Instruct
$1.72/M tokens
cheapest we can explain, at Alibaba Cloud (China) · $0.224 per 1,000 pages
131k context · tool calling · open weights
List price in (lower wins)
$0.60/M
$0.70/M
List price out (lower wins)
$3/M
$2.80/M
Cheapest we can explain (lower wins)
$2.80/M at NanoGPT
$1.72/M at Alibaba Cloud (China)
Standard price (lower wins)
$4.80/M
$4.90/M
Cost per 1,000 pages (lower wins)
$0.294
$0.224
Context
262k tokens
131k tokens
Open weights
yes
yes
Tool calling
yes
yes
Structured output
yes
not stated
Reads images
yes
no
Reasoning
yes
no
Released
27 Jan 2026
1 Sept 2024
Providers selling it
52
4
Price we measure against
the maker's own
the maker's own
Trains on your prompts (no is better)
not stated
not stated
Providers selling both · 1
Alibaba Cloud (China)$4.13 / $1.72
Kimi K2.5 then Qwen2.5 32B Instruct, per million tokens
Go further
Add a third or fourth model, or switch the weighting, in the tool.
What moved in the last seven days
Frequently asked
- Which is cheaper, Kimi K2.5 or Qwen2.5 32B Instruct?
- Qwen2.5 32B Instruct, at $1.72 per million tokens against $2.80 for Kimi K2.5, three parts input to one part output, read on 16 Sept 2026.
- Where is each one cheapest?
- Kimi K2.5: $2.80 at NanoGPT. Qwen2.5 32B Instruct: $1.72 at Alibaba Cloud (China).
- Can I buy both from one provider?
- Yes. One provider sells both: Alibaba Cloud (China). The cheapest of them for the two together is Alibaba Cloud (China).
- Which holds more context?
- Kimi K2.5, at 262k tokens against 131k.
Prices exclude tax and are per million tokens, read 16 Sept 2026 from each provider's own catalog. The standard price blends three parts input to one part output; a page is 600 tokens in and 60 out. Bars are scaled to the larger value in each row.