Alibaba

Key Terms

• Bailian (Model Studio) - Alibaba Cloud's managed LLM platform. Provides API access, token billing, and subscription plans for Qwen and third-party models. Source: Aliyun – Models
• Coding Plan - a flat-rate monthly subscription (¥200/month Pro tier) that provides request-based access to coding models through CLI tools. Billed per request, not per token. Prohibited for API/automated use. Source: Aliyun – Coding Plan
• Token Plan (Team Edition) - a per-seat monthly subscription with Credits-based billing. Three tiers: Standard (¥198/seat/month, 25K Credits), Advanced (¥698, 100K Credits), Premium (¥1,398, 250K Credits). Consumption depends on model, tokens, thinking mode, and tool calls. Source: Aliyun – Token Plan Overview
• Credits - Token Plan's billing unit. A qwen3.6-plus request with ~8,349 input tokens, ~40,794 cached tokens, and ~573 output tokens consumes approximately 3.18 Credits. Source: Aliyun – Token Plan Overview
• Context caching - reusing previously cached input tokens at a discount. Supported on qwen3.7-max, qwen3.7-plus, and other models. Reduces input token cost. Source: Aliyun – Context Cache
• Batch API - asynchronous inference at 50% of real-time pricing. Supported on qwen3.7-max, qwen3.7-plus, and other models. Source: Aliyun – Batch Interfaces Compatible With Openai
• Tiered pricing - input token cost rises with larger single-request context. For example, qwen3.7-plus costs ¥2/MTok input for requests up to 256K tokens, rising to ¥6/MTok for 256K-1M tokens. Source: Aliyun – Billing For Model Studio
• Implicit caching - automatic context caching on the Token Plan without an explicit cache call. Enabled for qwen3.7-max during the promotional period. Source: Aliyun – Token Plan Overview

Latest Changes

Changes since the 2026-05 report.

• Verified (May watch-item): qwen3.7-max Token Plan promo did NOT end June 22. Alibaba extended the promotional period to 2026-07-22 23:59 (UTC+8). Credits consumption for qwen3.7-max remains halved, and implicit caching is still supported on the Token Plan. This is a one-month extension of the promo May's report expected to expire. Source: Aliyun – Token Plan Overview
• New model: qwen3.7-plus launched 2026-06-01 (snapshot qwen3.7-plus-2026-05-26). A multimodal successor to qwen3.6-plus that adds visual-language understanding, screen reading, GUI operation, and visual-reference code generation, while keeping full coding, tool-use, and productivity agent capability. Listed as a top-tier Qwen text model on the catalog. Source: Aliyun – Newly Released Models
• Price drop (output): qwen3.7-plus is cheaper than qwen3.6-plus. At ≤256K tokens, qwen3.7-plus is ¥2/MTok input and ¥8/MTok output (thinking and non-thinking both ¥8), versus qwen3.6-plus at ¥2/MTok input and ¥12/MTok output. At 256K-1M tokens, qwen3.7-plus is ¥6/¥24 versus qwen3.6-plus at ¥8/¥48. A roughly 33% output cut at the lower tier. Source: Aliyun – Billing For Model Studio
• New snapshot: qwen3.7-max-2026-06-08 (added 2026-06-10). Adds visual-modality understanding to the Max flagship (the 2026-05-20 snapshot was text-only), enabling multimodal interactive agent capability. Same price (¥12/¥36 per MTok). Note: the bare alias qwen3.7-max still resolves to the 2026-05-20 snapshot, so the vision variant must be called by its explicit dated ID. Source: Aliyun – Newly Released Models
• New third-party model: kimi/kimi-k2.7-code (added 2026-06-15). Moonshot direct-supply, coding-centric agent model optimized for long-horizon software engineering (cross-file refactors, long debugging sessions), thinking mode only. A high-speed variant kimi-k2.7-code-highspeed (added 2026-06-18) runs 5-6x faster at identical capability. Source: Aliyun – Newly Released Models
• New third-party model on marketplace: glm-5.2. Now listed on the model catalog and Token Plan support list, superseding glm-5.1 as the top Zhipu model on Bailian. Source: Aliyun – Models
• New third-party model on marketplace: MiniMax-M3. Now listed on the model catalog text-generation section, though the Token Plan supported-model list still shows MiniMax-M2.5 (M3 not yet on Token Plan). Source: Aliyun – Models
• Catalog reshuffle: qwen3.6-plus demoted. The model catalog's headline Qwen trio is now qwen3.7-max, qwen3.7-plus, qwen3.6-flash. qwen3.6-plus remains available (pay-as-you-go and Token Plan) but is no longer the featured Plus model. Source: Aliyun – Models
• HLD update needed: HLD.md lists the Latest LLM as "qwen3.7-max, qwen3.6-plus, qwen3.6-flash" but the current featured Plus model is qwen3.7-plus. The Latest LLM column should change qwen3.6-plus to qwen3.7-plus.

Plans

Terms explained:

• Credits - Token Plan's billing unit. A single qwen3.6-plus request with ~8K input tokens consumes roughly 3.18 Credits. Actual consumption varies by model, thinking mode, and tool usage. Source: Aliyun – Token Plan Overview
• Data used for training - Token Plan explicitly states it does not use conversation data to train models. The Coding Plan explicitly grants Alibaba a license to use inputs and outputs during the subscription. This distinction is critical for enterprise buyers. Source: Aliyun – Token Plan Overview
• Interactive-only use (Token Plan) - the Token Plan terms restrict use to "interactive use in compatible AI programming and agent tools" and prohibit use for "automated scripts or application backends." Violation can pause the subscription or ban the API key. Source: Aliyun – Token Plan Overview

Source: Aliyun – Token Plan Overview

API Pricing

All prices in CNY per million tokens unless noted. China mainland region (Beijing) is the headline rate; international regions differ (noted below). Approximate USD at ~$1 = ¥7.2.

Qwen3.7 Max (flagship)

Features: Batch API (50% discount), context caching. Free tier: 1M tokens input + 1M output, 90-day expiry. 1M context window, 64K max output, 256K thinking budget (per May release notes). Source: Aliyun – Billing For Model Studio

Qwen3.7 Plus (new, multimodal, launched 2026-06-01)

Features: Batch API (50%), context caching. Free tier: 1M tokens each input/output, 90-day expiry. 1M context window. Multimodal: text, vision, GUI operation, screen reading. Source: Aliyun – Billing For Model Studio

Qwen3.6 Plus (predecessor, still available)

Features: Batch API (50%). 1M context window. Source: Aliyun – Billing For Model Studio

Qwen3.6 Flash (budget tier, unchanged)

Features: Batch API (50%), context caching. 1M context window. Note: flash tier pricing is carried forward from the 2026-05 report; the live billing page was truncated before the Flash section rendered, so it was not re-verified this month. Source: Aliyun – Billing For Model Studio

Regional price differences (qwen3.7-max)

Source: Aliyun – Billing For Model Studio

Approximate USD Equivalent (Beijing rate, at $1 = ¥7.2)

Model Performance / Benchmarks

Alibaba has not published coding-specific benchmark scores (SWE-Bench Verified, TerminalBench, LiveCodeBench, etc.) for any closed-source model: qwen3.7-max, qwen3.7-plus, qwen3.6-plus, or qwen3.6-flash. The release notes describe capabilities qualitatively ("excels in programming, office productivity, and long-horizon autonomous execution") without numbers. Source: Aliyun – Newly Released Models

The only verifiable, community-scored Qwen data points are for the open-weight Qwen3.6 models, via the Artificial Analysis intelligence index as reported in an independent June blog post:

These are third-party scores on the open models, not Alibaba-published numbers for the API-served closed models. Source: Quesma – Qwen 36 Is Awesome

Latest News

qwen3.7-plus Launched as Multimodal Plus Model (2026-06-01)

Alibaba released qwen3.7-plus (snapshot 2026-05-26), positioning it as the successor to qwen3.6-plus. It upgrades visual-language capability on top of strong text ability while retaining full coding, tool-use, and productivity agent behavior. Its headline feature is "multimodal interactive mixed agent capability": perceiving real-world scenes, reading screens and operating GUIs, generating code from visual references, and end-to-end mobile-app navigation. Source: Aliyun – Newly Released Models

qwen3.7-max Gets a Vision Snapshot (2026-06-10)

A new snapshot qwen3.7-max-2026-06-08 adds visual-modality understanding to the Max flagship (the May 20 snapshot was text-only), described as adding multimodal interactive agent capability. Pricing is unchanged at ¥12/¥36 per MTok. The bare qwen3.7-max alias still points to the 2026-05-20 snapshot, so the vision variant is opt-in via the explicit dated ID. Source: Aliyun – Newly Released Models

Token Plan Promo Extended to July 22 (ongoing)

The qwen3.7-max half-Credits promotional period, which May's report expected to end June 22, was extended to 2026-07-22 23:59 (UTC+8). Credits consumption for qwen3.7-max on the Token Plan is halved, and implicit caching is supported. Source: Aliyun – Token Plan Overview

Kimi K2.7 Code Added, Plus a High-Speed Variant (2026-06-15 / 2026-06-18)

Moonshot's kimi/kimi-k2.7-code, a coding-centric agent model optimized for long-horizon software engineering (thinking mode only), was added on June 15. A high-speed variant kimi-k2.7-code-highspeed followed on June 18, offering identical capability at 5-6x the speed. These are Moonshot direct-supply, not Qwen models, but expand the Bailian coding lineup. Source: Aliyun – Newly Released Models

Third-Party Model Refresh (ongoing)

glm-5.2 (Zhipu) and MiniMax-M3 now appear on the model catalog text-generation section, superseding glm-5.1 and MiniMax-M2.5 as the newest from those vendors. glm-5.2 is also on the Token Plan support list; MiniMax-M3 is marketplace-only and not yet on Token Plan (which still lists MiniMax-M2.5). Source: Aliyun – Models

Qwen Blog Still Redirects to qwen.ai

The old blog at qwenlm.github.io/blog/ continues to redirect to qwen.ai/research. The last substantive post on the old blog remains "Qwen3Guard" (September 23, 2025). The new qwen.ai location is heavily JavaScript-rendered and no June 2026 posts were extractable via standard fetch. Model-release detail is therefore sourced from the Aliyun help docs rather than the Qwen blog. Source: Alibaba

Community Signals

"Qwen 3.6 27B is the sweet spot for local development" hit the HackerNews front page (June 29)

An independent blog post arguing the open-weight Qwen 3.6 27B dense model is the first local model that "actually makes sense as a general intelligence" reached the front page, accumulating 1,161 upvotes and 723 comments. The author benchmarks it at 32 tokens/s with multi-token prediction on a 128GB MacBook, scores it at 37 on the Artificial Analysis index (mid-2025 frontier level), and runs it through OpenCode for real vibe-coding tasks (a hexagonal minesweeper built from a single prompt). Source: News – Item , Quesma – Qwen 36 Is Awesome

Developer quotes on using local Qwen 3.6 27B for real work

• ZamByte: "I have been using pi (and previously the codex cli) with Qwen 3.6 27b with 100k context for my development at work, and I have been very blown away by how well it works. It's not perfect, but it's enough to accelerate my normal development flow. I mostly use it for writing Go and C#." News – Item
• mr_mitm, on commercial drivers for local Qwen: "Think commercial. My company invested in a local rig since privacy is important to our customers and sometimes I want to use these models on private data." News – Item
• jboss10, on hardware feasibility: "A 27B model can fit easily on a 32GB VRAM card (e.g. 5090) or a 32GB computer in RAM at FP8/Q8 (unsloth have 28.6GB Q8 files). For 24GB VRAM cards (e.g. 4090) you can use Q6_K (22.5GB) or Q5_K_M (19.5GB) quants, possibly offloading some of the weights to RAM." News – Item
• sleepyeldrazi: "I've been running it almost since launch on a 3090 (24gb vram), you really don't need that much. Second hand those are really cheap and i get 50-70 t[okens/s]." News – Item

Skeptical counterpoint on Qwen vs US frontier models

Not all feedback was positive. One developer argued Qwen's benchmark strength masks weaker generalization relative to US frontier models: "This is my experience too. Qwen optimizes for a lot of scenarios which masks their weaker generalization compared to US frontier models. Never go below an fp[8 quant]." News – Item

Memory-bandwidth and frontier-context debate

Simon Willison (simonw) contextualized the local-Qwen excitement against the gap to the largest open models: "It can't run the latest models today - GLM-5.2 class models already need 1TB+ of RAM. ... but, the models that WILL run on 128GB (or 64GB or even 32GB) [will improve]." News – Item

Enterprise Readiness

Terms explained:

• Data residency - pay-as-you-go is multi-region, but the subscription plans (Coding Plan, Token Plan) are Beijing-only. A team requiring data processing outside China cannot use the seat-based plans. Source: Aliyun – Token Plan Overview
• IP indemnity - no intellectual property indemnification is offered for Qwen outputs, unlike Microsoft's Copilot Copyright Commitment or comparable programs.

Transparency Gaps