Qwen 3.7 Plus

The Overchat AI article, last updated 2026-09-08, covers the Qwen 3.7 family and confirms key Qwen 3.7 Plus attributes in passing: Plus is multimodal and can understand images, distinguishing it from the text-only Qwen 3.7 Max. The page states that Qwen 3.7 Max was launched on 20 May 2026 at the Alibaba Cloud Summit, w The article corroborates the pricing tier for the Max variant at $2.50 per million input tokens and $7.50 per million output tokens and notes Anthropic API compatibility, which lets Qwen 3.7 Max work with Claude Code directly. For the Plus variant specifically, the page supplies only the modality distinction (multimoda

Alibaba's Qwen team released Qwen3.7-Plus on June 2, 2026, as a multimodal agent model generally available through Alibaba Cloud's Bailian platform (marketed internationally as Model Studio) via API. The model accepts text, images, and video as input but outputs only text, combining visual perception with GUI operation On Alibaba-reported benchmarks, Qwen3.7-Plus leads on ScreenSpot Pro (79.0 versus GPT-5.4's 67.4) and AndroidWorld (81.0 versus Gemini 3.1 Pro's 70.7), though it trails on pure reasoning tasks. API pricing is set at $0.40 per million input tokens and $1.60 per million output tokens — roughly six times cheaper than Qwen

BenchGecko aggregates Qwen3.7 Plus by Alibaba as a multimodal, open-source model released in June 2026 with a 1.0M-token context window. The listing reports input pricing of $0.32 per 1M tokens and output pricing of $1.28 per 1M tokens, placing it as the cost-effective tier within Alibaba's Qwen 3.7 family. It shows benchmark coverage on three Artificial Analysis composite indices: Coding Index 55.9%, Quality Index 39.0%, and Agentic Index 20.8%, giving developers a snapshot of where Plus sits relative to broader model rankings. The page situates Qwen3.7 Plus within the Qwen 3.7 family lineup, contrasting it with the more expensive Qwen3.7 Max (May 2026, $1.25/M input, 1.0M context, 18 benchmarks tested) and listing similar models from other providers for quick comparison. It also links to the model's technical report, documentation, API docs, playground, GitHub, and Hugging Face pages, serving as a developer-oriented index entry. Specifications are summarized as multimodal type, 1.0M-token context, "Open Source" license status, and active availability across listed providers.

According to ofox.ai's comparison article (published 2026-06-02), Alibaba shipped Qwen 3.7 Plus on June 1, 2026, eleven days after Qwen 3.7 Max launched on May 21, 2026. The page supplies concrete variant-level facts for Qwen 3.7 Plus: a 1,000,000-token context window, multimodal support for text, image, and video, pri The same page frames Qwen 3.7 Plus as a value-tier sibling to Max, roughly 6x cheaper across input, output, and cached tokens while retaining the same 1M context window and 35-hour autonomous ceiling, and it notes that both models ship through Alibaba's Bailian platform alongside ofox's OpenAI-compatible endpoint. Per

Vals.ai profiles Qwen 3.7 Plus as a model from Alibaba released June 1, 2026, with a 1M-token context window, 66k max output tokens, and token costs of $0.40 input / $1.60 output per 1M tokens. Input modalities are listed as text, image, and video supported (file input not supported), with weights marked as private. The page reports a Vals Index accuracy of 38.65% ±1.17 at a cost per test of $0.440 and latency of 21 min 48 s, alongside per-benchmark deltas across Code Migration, EMB, Finance Agent v2, Legal Research Bench, MortgageTax, SAGE, Vibe Code Bench v1.1, Harvey's Legal Agent Benchmark, SkillsBench, and Terminal-Bench 2.1. The platform also tracks an update dated June 2, 2026, noting that Alibaba's Qwen 3.7 Plus was evaluated on the Vals Index, where it ranked 13 with a score of 52.33%, with its strongest component being Vibe Code Bench (ranking 8 on the subset). Default provider settings are listed as Alibaba with temperature 0.7, default top P/top K, and max output tokens 65,536, providing a reference configuration for reproducibility of the benchmark runs.

Artificial Analysis benchmarks Qwen3.7 Plus at the API-provider level and lists Alibaba Cloud as the single provider for the model. Reported performance metrics include 73.5 tokens/second output speed, 29.49 seconds time-to-first-token, and a blended price of $0.30 per 1M tokens. The page breaks down cache hit, input, and output pricing in USD per million tokens and notes that the default benchmarking workload has been updated to 10k input tokens to better reflect production agentic use cases. The analysis supports provider-selection decisions for Qwen3.7 Plus by combining price-vs-speed visualizations and a cache discount metric defined as 1 − (cache hit price / input price). These figures give developers concrete numbers for choosing between serving providers, alongside cache-pricing details and a recommended 7:2:1 cache-input-output blending ratio for general agentic workloads. Only one provider is currently profiled, with additional notes available via the underlying cache-pricing detail page.

Alibaba Cloud's official community blog introduces Qwen3.7-Plus as a multimodal agent model that unifies vision and language into a single agent foundation. Built on Qwen3.7's text backbone, the model delivers an upgrade in vision-language capabilities while retaining agentic strength in coding, tool use, and productivity workflows. It is positioned as a "multimodal interactive hybrid agent" that can perceive real-world scenes, read screens and operate GUIs, write code from visual references, navigate mobile apps end-to-end, and answer visual questions grounded in web knowledge — blending GUI and CLI interactions within a single agent loop. The post confirms availability via Alibaba Cloud Model Studio and notes cross-harness generalization, allowing deployment through frameworks such as Claude Code, OpenClaw, or Qwen Code. The blog also details benchmark methodology and results, including Terminal-Bench 2.0 runs (Harbor/Terminus-2 harness, 5h timeout, 12 CPU/24 GB RAM, temperature 1.0, top_p 0.95, top_k 20, max_tokens 80K, 256K context, average of 5 runs with an extended-thinking token prepended per turn), SWE-Bench Series results using an internal agent scaffold with bash and file-edit tools at 200K context, and a refined SWE-bench Pro evaluation. It cites a custom QwenClawBench built on a real-user-distribution Claw agent distribution. Combined, these specifics confirm the model is a multimodal agent with tool use, GUI/CLI hybrid operation, and full-modality input support, available through the first-party Alibaba Cloud Model Studio API.