Which Qwen3.5 Plus conclusions hold up?

Qwen3.5-Plus: Qwen's Official Native Multimodal Agent Release Baseline

It positions Qwen3.5 Plus as the hosted API counterpart to Qwen3.5-397B-A17B, covering vision, multimodal agents, tool calls, and a million-token context; the tables remain bounded by Alibaba’s harness.

Source-specific observation

The February 15, 2026 Qwen release compares Qwen3.5 Plus and Qwen3.5-397B-A17B across language, coding, agent, and multimodal tasks.

Published conditions

Prompts, sampling, tool schemas, repeats, and the full benchmark harness are not public; the Model Studio page separately lists text/image/video, function calling, and structured output.

Qwen3.5-Plus: Qubrid's Same-Image, Same-Prompt Latency and Token Comparison

Qubrid records latency, token allocation, and output differences in one same-image, same-prompt run; the backend snapshot is undisclosed, so the result is Playground-specific.

Unverified: the original source could not be rechecked.

Source-specific observation

Qubrid compares Qwen3.5 Plus and Qwen3.6 Plus with the same image and prompt and records TTFT, total time, tokens/s, and token categories.

Published conditions

It reports 1,858 reasoning tokens out of 2,036 completion tokens for Qwen3.5 Plus; provider snapshot, repeats, and the full harness are unpublished.

Qwen3.5-Plus: Digital Applied's Cross-Model Benchmarks and Hosted/Open-Weight Selection

Digital Applied summarizes Qwen3.5 benchmarks, API migration, and hosted-versus-open-weight tradeoffs; it has no uniform item-level harness and is best treated as an independent synthesis.

Unverified: the original source could not be rechecked.

Source-specific observation

The February 16, 2026 guide discusses Qwen3.5 Plus and Qwen3.5-397B-A17B and shows an OpenAI-compatible SDK example.

Published conditions

Item prompts, seeds, repeats, and a common harness are not published; some figures are vendor or secondary data.

Qwen3.5-Plus: Qwen's Official Native Multimodal Agent Release Baseline

It positions Qwen3.5 Plus as the hosted API counterpart to Qwen3.5-397B-A17B, covering vision, multimodal agents, tool calls, and a million-token context; the tables remain bounded by Alibaba’s harness.

Source-specific observation

The February 15, 2026 Qwen release compares Qwen3.5 Plus and Qwen3.5-397B-A17B across language, coding, agent, and multimodal tasks.

Published conditions

Prompts, sampling, tool schemas, repeats, and the full benchmark harness are not public; the Model Studio page separately lists text/image/video, function calling, and structured output.

Qwen3.5-Plus: Qubrid's Same-Image, Same-Prompt Latency and Token Comparison

Qubrid records latency, token allocation, and output differences in one same-image, same-prompt run; the backend snapshot is undisclosed, so the result is Playground-specific.

Unverified: the original source could not be rechecked.

Source-specific observation

Qubrid compares Qwen3.5 Plus and Qwen3.6 Plus with the same image and prompt and records TTFT, total time, tokens/s, and token categories.

Published conditions

It reports 1,858 reasoning tokens out of 2,036 completion tokens for Qwen3.5 Plus; provider snapshot, repeats, and the full harness are unpublished.

Qwen3.5-Plus: Digital Applied's Cross-Model Benchmarks and Hosted/Open-Weight Selection

Digital Applied summarizes Qwen3.5 benchmarks, API migration, and hosted-versus-open-weight tradeoffs; it has no uniform item-level harness and is best treated as an independent synthesis.

Unverified: the original source could not be rechecked.

Source-specific observation

The February 16, 2026 guide discusses Qwen3.5 Plus and Qwen3.5-397B-A17B and shows an OpenAI-compatible SDK example.

Published conditions

Item prompts, seeds, repeats, and a common harness are not published; some figures are vendor or secondary data.