Qwen3.5 397B-A17B vs Trinity Large Thinking vs GLM-5

It is close. Our weighted score puts them within 1 points (Qwen3.5 397B-A17B 56/100, Trinity Large Thinking 55/100, GLM-5 37/100), so choose by what matters most for your work: Trinity Large Thinking on price and Trinity Large Thinking for long inputs. The score weighs price 50%, inputs & features 30%, context window 20%. None of the shared benchmarks cover every model here yet, so capability is left out of this verdict.

Trinity Large Thinking is cheaper at $0.25 input / $0.80 output per million tokens (official Arcee API price). Qwen3.5 397B-A17B costs $0.60 input / $3.60 output per million tokens (official Alibaba API price); GLM-5 costs $1.00 input / $3.20 output per million tokens (official Z.AI API price). At a typical mix of three input tokens to one output token, that is $0.388 per million tokens for Trinity Large Thinking versus $1.35 for Qwen3.5 397B-A17B (3.5× as much) and $1.55 for GLM-5 (4× as much).

There is no independent benchmark that covers all three models yet. Qwen3.5 397B-A17B has an ECI of 146.7, Trinity Large Thinking has not been scored yet and GLM-5 has an ECI of 145.8.

There are no published SWE-bench Verified results for Qwen3.5 397B-A17B and Trinity Large Thinking yet, so there is no like-for-like coding score. Test both on a sample of your own tasks. All three support tool calling for agent workflows.

Trinity Large Thinking has the largest context window at 524,288 tokens, against 262,144 for Qwen3.5 397B-A17B and 204,800 for GLM-5. Maximum output per response: Qwen3.5 397B-A17B up to 65,536, Trinity Large Thinking up to 262,144, GLM-5 up to 131,072 tokens.

Qwen3.5 397B-A17B accepts text, images, audio and video; Trinity Large Thinking accepts text; GLM-5 accepts text. Qwen3.5 397B-A17B handles the widest range of inputs.

Yes, all three publish their weights (OpenMDW-1.1), so you can self-host them.

Trinity Large Thinking is the newest, released Apr 1, 2026. Qwen3.5 397B-A17B came out Feb 15, 2026; GLM-5 came out Feb 12, 2026.

Each model gets a 0–100 score on capability (50%, independent benchmark results); price (25%, blended price per million tokens (3 input : 1 output), log scale); inputs & features (15%, image, PDF, audio and video input, tool calling, structured output and reasoning); context window (10%, maximum tokens per request, log scale). Dimensions missing for any model are dropped and the remaining weights rescaled, so every model is judged on the same evidence. Specs and prices come from public model listings and the labs’ own API pages; capability scores come from independent benchmark runs. Data updated Oct 4, 2026.