Qwen3.8 27B benchmarks
Qwen3.8 27B is an open-weights model from Alibaba, released 14 Aug 2026. It ranks 46th of 51 on our overall leaderboard and 13th of 51 for long context. At A$1.58 per million tokens (blended), it costs about 2.3× cheaper than the median model we track. It writes about 46 tokens a second, and answers start after 45 s on average, thinking included. Its weights are public, so it can be hosted in Australia on your own infrastructure or an Australian cloud region.
- CONTEXT WINDOW
- 1,000,000 tokens
- MAX OUTPUT
- 131,072 tokens
- INPUT
- text, image, video
- WEIGHTS
- Open (Apache 2.0)
- RELEASED
- 14 Aug 2026
- OUTPUT SPEED
- 46 tok/s
- ANSWER STARTS AFTER
- 45 s (thinking included)
Qwen3.8 27B across our rankings.
Benchmark results.
Independent test scores are for the xhigh effort setting. “Among tracked” ranks Qwen3.8 27B against the 51 models on this site; LMArena ranks run across its full leaderboard.
| BENCHMARK | RESULT | AMONG TRACKED | SOURCE DETAIL |
|---|---|---|---|
| INDEPENDENT TESTS · ARTIFICIAL ANALYSIS | |||
| AA Intelligence Index | 33.7 | #34 of 51 | xhigh effort |
| AA Coding Index | 68.1 | #31 of 42 | xhigh effort |
| GPQA Diamond | 90.5% | #32 of 44 | xhigh effort |
| Humanity’s Last Exam | 33.9% | #45 of 51 | xhigh effort |
| AA-LCR | 82.0% | #16 of 51 | xhigh effort |
| Terminal-Bench 2.1 | 79.8% | #23 of 42 | xhigh effort |
| τ²-Bench (banking) | 48.0% | #4 of 41 | xhigh effort |
| SciCode | 46.6% | #43 of 46 | xhigh effort |
| BLIND HUMAN VOTES · LMARENA | |||
| LMArena Text10,697 votes · ±7 | 1437 | #41 of 42 | #90 of 402 on LMArena · 13 Sep 2026 |
| LMArena WebDev9,092 votes · ±8 | 1591 | #17 of 45 | #21 of 129 on LMArena · 22 Sep 2026 |
| LMArena Vision4,577 votes · ±10 | 1244 | #31 of 33 | #49 of 152 on LMArena · 13 Sep 2026 |
| LMArena Agent37,257 sessions | #29 | #26 of 37 | #29 of 46 on LMArena · 15 Sep 2026 |
What Qwen3.8 27B costs in Australian dollars.
List prices converted at the snapshot’s RBA rate, excluding GST. Reasoning models are billed for their thinking as output tokens, so heavier settings cost more than the job estimates show.
| PER MILLION TOKENS | AUD | USD LIST |
|---|---|---|
| Input tokens | A$0.70 | US$0.50 |
| Output tokens | A$4.21 | US$3.00 |
| Blended (3 in : 1 out) | A$1.58 | US$1.13 |
| JOB | COST (AUD) | MEDIAN MODEL |
|---|---|---|
| Customer support replyPER 1,000 REPLIES | A$2.67 | ≈ A$6.58 |
| Summarise a 30-page documentPER 100 DOCUMENTS | A$1.74 | ≈ A$4.74 |
| Agentic coding taskPER 10 TASKS | A$1.39 | ≈ A$3.72 |
| SETTING | INTELLIGENCE | A$ / 1M | SPEED |
|---|---|---|---|
| xhigh effort | 33.7 | A$1.58 | 46 tok/s |
| medium effort | 27.6 | A$1.58 | 47 tok/s |
| low effort | 26.2 | A$1.58 | 52 tok/s |
| Non-reasoning | 20.2 | A$1.58 | 49 tok/s |
Running Qwen3.8 27B in Australia.
Its weights are public, so it can be hosted in Australia on your own infrastructure or an Australian cloud region. How the platforms compare →
| PLATFORM | AVAILABILITY | DETAIL |
|---|---|---|
| AWS Bedrock · Sydney | NOT OFFERED | |
| AWS Bedrock · Melbourne | NOT OFFERED | |
| Azure · Australia East | NOT OFFERED | |
| Google Vertex AI · Sydney | NOT OFFERED | |
| Your own infrastructure | SELF-HOST | Open weights: run it on your own servers or GPU instances in an Australian region. |
Compare it with.
Qwen3.8 27B, answered.
How much does Qwen3.8 27B cost in Australian dollars?
Qwen3.8 27B lists at US$0.50 per million input tokens and US$3.00 per million output tokens — A$0.70 and A$4.21 at A$1 = US$0.7123 (RBA, 22 Sep 2026), excluding GST. A typical customer support reply works out at about A$2.67 per 1,000 replies, before any reasoning tokens.
Can I use Qwen3.8 27B in Australia with data kept onshore?
Qwen3.8 27B’s weights are public, so it can be hosted in Australia on your own infrastructure or an Australian cloud region. Availability comes from the cloud providers’ own documentation; check your provider’s current terms before relying on it for data-residency obligations.
Is Qwen3.8 27B good for coding?
It ranks 26th of 42 on our coding ranking, with an Artificial Analysis Coding Index of 68.1 and an LMArena WebDev rating of 1591 (21st of 129). The current leader is Claude Fable 5.1.
How good is Qwen3.8 27B at agent and automation work?
It ranks 14th of 41 on our agents ranking, completing 48.0% of τ²-Bench banking customer-service tasks and 79.8% of Terminal-Bench 2.1 tasks.
How fast is Qwen3.8 27B?
Artificial Analysis measures Qwen3.8 27B at about 46 tok/s of output, with the answer starting after 45 s on average once thinking time is included (xhigh effort setting).
What is Qwen3.8 27B’s context window?
Alibaba documents a context window of 1,000,000 tokens (1M), with up to 131,072 output tokens. On AA-LCR, which tests reasoning across ~100,000-token document sets, it scores 82.0%.
Ratings: LMArena leaderboard dataset (CC BY 4.0), rescaled for the lens scores · leaderboards to 22 Sep 2026
Evaluations, prices and speed: Artificial Analysis (artificialanalysis.ai) · fetched 23 Sep 2026
Exchange rate: Reserve Bank of Australia, table F11.1 · A$1 = US$0.7123 on 22 Sep 2026 · prices exclude GST
Snapshot 23 Sep 2026 · updated weekly · How the rankings work →