# Qwen 3.8-Max Preview — Investigation Summary ## What Happened On **July 19, 2026**, Alibaba's Qwen team announced **Qwen 3.8**, a new flagship model family, and immediately released **Qwen3.8-Max-Preview** via API. The announcement came via the official @Alibaba_Qwen X account, generating **1.4M+ views** in its first hours. ### Key Specs | Attribute | Detail | |---|---| | **Parameters** | 2.4 trillion (2.4T) — likely MoE architecture | | **Release Date** | July 19, 2026 | | **Availability** | API preview via Token Plan, Qoder, QoderWork | | **Open Weights** | Announced as "coming soon" — not yet downloadable | | **Context Window** | Not yet published | | **Modality** | Text → Text (reasoning model) | | **Focus Areas** | Complex reasoning, code engineering, professional office, long-range Agent tasks | ### What Changed from Qwen 3.7-Max | Dimension | Qwen 3.7-Max (May 2026) | Qwen 3.8-Max-Preview (July 2026) | |---|---|---| | **Scale** | Unknown parameter count (1M context) | 2.4T parameters (massive jump) | | **Pricing Model** | Per-token: $2.50/M input, $7.50/M output | Token Plan subscription: from **$6/mo** (Lite) | | **Open Weights** | Never opened | Promised "soon" | | **Positioning** | Agent-centric, programming, office | Code engineering + professional office + long-range agents | | **Claimed Rank** | Led Terminal Bench 2.0 (69.7) | "Second only to Fable 5" | | **Protocol Support** | OpenAI-compatible | OpenAI + Anthropic protocol endpoints | ### Token Plan Pricing (Individual) | Plan | Price | Credits | Concurrent Agents | |---|---|---|---| | **Lite** | **$6/mo** (launch discount from $8) | 2,500 weekly | 1–2 | | **Standard** | Higher tier | 4× Lite credits | 3–4 | | **Pro** | Highest tier | 16× Lite credits | 6–8 | The Token Plan bundles access to multiple frontier models: **qwen3.7-max, Qwen3.8-Max-Preview, glm-5.2, deepseek-v4-pro, wan2.7-image-pro** — all on one subscription. ### Supported Tools Works with tools supporting OpenAI and Anthropic protocols: **Qwen Code, Cline, Claude Code, Cursor, OpenCode, Codex, Kilo CLI, OpenClaw**. --- ## Why It Matters 1. **Frontier Model Escalation**: 2.4T parameters makes this one of the largest publicly accessible models. The timing — same day Anthropic's Fable 5 promo extension expires — is a deliberate competitive move. 2. **Subscription vs. Per-Token**: The Token Plan model ($6/mo for Lite) is a radical departure from per-token pricing. For heavy users, this undercuts Western frontier model costs by a wide margin. 3. **Open-Weight Promise**: If Alibaba follows through on open weights, this would be the largest open-weight model ever released — a potential democratization moment for the AI industry. 4. **Protocol Compatibility**: Native support for both OpenAI and Anthropic protocols means developers can drop Qwen3.8-Max-Preview into existing agent harnesses (Claude Code, Cursor, etc.) without code changes. 5. **"Second Only to Fable 5" Claim**: This is unverified — no benchmark tables were published. However, Qwen 3.7-Max already led Terminal Bench 2.0, so the claim is directionally plausible. Independent evals are needed. --- ## Strongest Sources (Cited) | Source | Type | Key Content | |---|---|---| | **@Alibaba_Qwen on X** (July 19, 2026) | Official announcement | 2.4T params, open-weight soon, "second only to Fable 5" | | **QwenCloud Token Plan page** (qwencloud.com/pricing/token-plan) | Official pricing | Token Plan tiers, model list, tool compatibility | | **DataLearnerAI model card** (datalearner.com) | Third-party model registry | Specs, architecture notes, limitations (no benchmarks yet) | | **explainx.ai analysis** (Yash Thakker, July 19, 2026) | Independent analysis | Pricing breakdown, Fable 5 claim reality check, setup guide | | **Qoder.com** | Official product page | Qoder integration, agentic coding features | | **QwenCloud Qwen3.7-Max page** | Official comparison baseline | Prior-gen specs, pricing, context limits | **Bottom line**: Qwen3.8-Max-Preview is a major escalation in the frontier AI race — massive scale (2.4T), aggressive subscription pricing ($6/mo), and a promise of open weights. The "second only to Fable 5" claim is unverified but plausible. Developers can try it today via the Token Plan with zero code changes to existing agent setups.