Alibaba Qwenopen weightsQwen3.8 27B pricing
$0.24 per 1M input tokens and $2.20 per 1M output tokens. In real work that is 0.35¢ per chat answer and 0.90¢ to summarize a 20-page PDF.
What Qwen3.8 27B costs per task
Token counts estimated from a measured model of the same lab (Qwen3.8 Flash), rescaled by tokenizer. Reasoning tokens are billed as output.
| Task | Cost | Tokens used | Price rank | Basis |
|---|---|---|---|---|
| Answer a chat question | 0.35¢$3.54 / 1,000 | 88 in · 689 out · 909 reasoning | 206 of 336 | estimated |
| Reply to a customer email | 0.16¢$1.59 / 1,000 | 326 in · 98 out · 587 reasoning | 209 of 336 | estimated |
| Write a 1,500-word blog post | ≥3.5¢$35.23 / 1,000 | 104 in · 0 out · 16,000 reasoning | 283 of 336 | estimated |
| Write 10 product descriptions | ≥3.5¢$35.28 / 1,000 | 345 in · 0 out · 16,000 reasoning | 293 of 336 | estimated |
| Translate 1,000 words | 0.32¢$3.20 / 1,000 | 1,386 in · 1,166 out · 136 reasoning | 129 of 336 | estimated |
| Proofread one page (500 words) | 0.42¢$4.21 / 1,000 | 777 in · 671 out · 1,159 reasoning | 216 of 336 | estimated |
| Summarize a 20-page PDF | 0.90¢$8.96 / 1,000 | 13,119 in · 501 out · 2,141 reasoning | 174 of 336 | estimated |
| Extract data from an invoice | 0.49¢$4.91 / 1,000 | 501 in · 561 out · 1,617 reasoning | 223 of 336 | estimated |
| Analyze a 200-row spreadsheet | ≥3.7¢$36.73 / 1,000 | 6,357 in · 0 out · 16,000 reasoning | 208 of 336 | estimated |
| Answer from 5 retrieved documents (RAG) | 0.72¢$7.25 / 1,000 | 3,751 in · 974 out · 1,912 reasoning | 210 of 336 | estimated |
| Classify 100 support tickets | 1.9¢$18.86 / 1,000 | 1,463 in · 534 out · 7,879 reasoning | 256 of 336 | estimated |
| Review 500 lines of code | ≥3.6¢$36.37 / 1,000 | 4,874 in · 0 out · 16,000 reasoning | 234 of 336 | estimated |
| Write a SQL query | 0.58¢$5.83 / 1,000 | 177 in · 405 out · 2,225 reasoning | 239 of 336 | estimated |
| Run a 50-step coding agent | 34¢$340 / 1,000 | 2,259,875 in · 15,150 out · 80,850 reasoning | 136 of 336 | modeled |
Price at every provider
List prices per 1M tokens (input / output) at each host serving Qwen3.8 27B through OpenRouter today. Hosts can differ in quantization, context and speed.
| Provider | Input / output | Cached input | Context | Uptime |
|---|---|---|---|---|
| T Tenstorrentfp4 | $0.02 / $0.50 | $0.009 | 262K | 100% |
 Chutesfp8 | $0.039 / $1.25 | $0.004 | 262K | 100% |
| N Near AIfp8 | $0.04 / $1.35 | $0.018 | 262K | 100% |
 Cerebrasfp16 | $0.99 / $1.49 | $0.99 | 65K | 100% |
 DeepInfrabf16 | $0.15 / $1.88 | $0.037 | 262K | 100% |
| A AkashMLfp8 | $0.23 / $1.98 | $0.05 | 262K | 100% |
| D Darkbloomfp4 | $0.035 / $2.20 | $0.018 | 262K | |
| R Rekafp8 | $0.04 / $2.20 | $0.02 | 262K | 100% |
 Parasailfp8 | $0.24 / $2.20 | $0.05 | 262K | 100% |
| W Waferfp8 | $0.04 / $2.30 | $0.039 | 262K | 100% |
| I Ionstreamfp8 | $0.089 / $2.35 | $0.085 | 262K | 100% |
| D DekaLLMunknown | $0.04 / $2.50 | $0.02 | 262K | 100% |
| M ModelRunfp4 | $0.20 / $2.50 | $0.09 | 262K | 100% |
| M Mancer 2fp8 | $0.20 / $2.50 | - | 262K | |
| P Phalaunknown | $0.20 / $2.50 | $0.05 | 1M | 100% |
 Alibabaunknown | $0.42 / $2.55 | $0.085 | 1M | 100% |
| C CoreWeavefp8 | $0.40 / $3.00 | $0.15 | 262K | 30% |
 Novitaunknown | $0.42 / $3.00 | $0.085 | 1M | 100% |
| C Cloudflareunknown | $0.45 / $3.20 | $0.05 | 262K | 100% |
| V Venicefp8 | $0.45 / $3.20 | - | 262K | 100% |
Price history
First seen in our price history on Aug 21, 2026 at $0.45 input / $3.20 output. Since then the output price moved from $3.20 to $2.20 per 1M tokens (7 changes).
FAQ
How much does Qwen3.8 27B cost?
Qwen3.8 27B costs $0.24 per 1 million input tokens and $2.20 per 1 million output tokens, with cached input at $0.05 (list price on Oct 18, 2026). In practice one chat answer costs about 0.35¢ and summarizing a 20-page PDF about 0.90¢.
Where is Qwen3.8 27B cheapest?
Among 20 providers we track, the lowest list price today is at Tenstorrent: $0.02 input and $0.50 output per 1M tokens.
Does Qwen3.8 27B charge for reasoning tokens?
Yes. Reasoning (thinking) tokens are billed as output tokens. That is why the cost per task can be much higher than the visible answer suggests - our task costs include them.
Is the Qwen3.8 27B cost per task measured or estimated?
Estimated: Qwen3.8 27B was not run directly. Tokens come from Qwen3.8 Flash and are rescaled by the tokenizer of Alibaba Qwen. Real usage can differ.