Throughput (tokens / sec)
avg. 105tokens/s
Dense agentic coding with thinking control
Run instantly. Pay only for what you use.
Qwen 3.6 27B is Alibaba's dense open-weight coding model under Apache 2.0: all 27B parameters active on every token, a 262K-token context window, and native multimodal input — text, images, and video. It was the first open model to ship thinking preservation, carrying reasoning traces across conversation turns.
Dense weights make its quality per parameter stand out: on agentic coding it beats MoE models more than 10× its size — 53.5 on SWE-bench Pro against 50.9 for Qwen3.5-397B-A17B — while hybrid linear attention and multi-token prediction keep serving efficient.
With thinking enabled it posts 77.2% on SWE-bench Verified and 59.3 on Terminal-Bench 2.0, with strict function calling, JSON output, and per-request thinking control to trade reasoning depth for speed.
27Bdense parameters
Every parameter is active on every token — consistent quality with no routing variance.
262Ktoken context window
Holds full repositories and long coding sessions in a single context.
1stopen model with thinking preservation
Reasoning traces persist across turns, cutting redundant thinking tokens in agent loops.
Qwen 3.6 27B is strongest where step quality matters more than raw throughput — repository-level coding, visual front-end work, and multi-turn agents that build on earlier reasoning.
77.2% on SWE-bench Verified and 59.3% on Terminal-Bench 2.0 with thinking enabled — plans, edits, and verifies real repositories.
53.5 on SWE-bench Pro — ahead of MoE models more than 10× its size, from 27B dense weights.
Chain-of-thought carries across conversation turns, reducing redundant reasoning and KV-cache pressure in multi-turn agents.
Screenshots, diagrams, and video frames are first-class inputs — front-end bugs and visual tickets stay in one model.
Repository-level coding agents and refactors
Resolves 77.2% of SWE-bench Verified issues with thinking enabled — consistent step quality with no expert-routing variance.
Front-end work from screenshots and recordings
Screenshots, UI recordings, and diagrams are native inputs, so visual bugs and design-to-code tasks stay in one model.
Multi-turn agent loops
Thinking preservation keeps chain-of-thought across turns, cutting redundant reasoning tokens and KV-cache pressure.
Multilingual engineering teams
71.3 on SWE-bench Multilingual — issue-to-patch work on non-English repositories and tickets.
Cost-first, high-volume agent fleets
Every token runs all 27B parameters. When per-token cost dominates, use Qwen 3.6 35B-A3B — near-peer quality at 3B-active pricing.
Tight output-token budgets with thinking on
Thinking mode is verbose — roughly 4× the median output tokens in Artificial Analysis testing. Cap or disable thinking for cost-sensitive replies.
Audio input or speech understanding
Inputs are text, image, and video only. Transcribe audio upstream before sending it to the model.
Compare Entrim’s pricing, powered by an optimized inference runtime, against other providers using the same token counts per request.
avg. 105tokens/s
avg. 780ms
77.2%SWE-bench Verified
Resolves real GitHub issues end-to-end in agentic coding runs.
53.5SWE-bench Pro
Harder, contamination-resistant software engineering tasks.
59.3Terminal-Bench 2.0
Autonomous work in a real terminal — builds, debugging, ops tasks.
Code examples
curl -X POST "https://api.entrim.ai/v1/chat/completions" \ -H "Content-Type: application/json" \ -H "Authorization: Bearer $ENTRIM_API_KEY" \ -d '{ "model": "Qwen/Qwen3.6-27B", "messages": [ { "role": "user", "content": "Why do they call it a building if it is already built?" } ] }'SDKs & docs
Use familiar SDK patterns with quickstart examples for common production setups.
OpenAI-compatible
Keep the request format your team already knows.
No cold starts
Production requests are served without model spin-up delays.
Tool-heavy coding at 3B-active cost
High-throughput agentic work with 1M context
Multimodal long-context work with thinking control