Qwen3.8 Max 0902

OpenRouter's listing names the exact model id qwen/qwen3.8-max-0902 and describes it as an updated snapshot of Qwen3.8 Max from Alibaba's Qwen team. It is a 2.4-trillion-parameter mixture-of-experts model accepting text, image, and video input and returning text, with a 1M-token context window, reasoning enabled by def Operational data shows Alibaba Cloud Int. as the sole host with $2/$6 per 1M tokens, $0.25 cache read, 39 tok/s P50 throughput, and 1.15s P50 latency. The page records a Sep 3, 2026 release date and an AutoExacto benchmark score of 63.6% attributed to Alibaba, while weighted-average effective input pricing lands at $0.

Alibaba's Qwen team launched Qwen3.8-Max-0902 on September 2, 2026, as a post-training upgrade of Qwen3.8-Max with the same base, focused on coding and cowork tasks. On release day, Arena.ai updated its Code Arena WebDev leaderboard, where the model took first place overall with 1691 points. That score edges Claude Opus 5 Max by 3 points, Kimi K3 Max by 17 points, and the prior Qwen3.8-Max by 22 points. The report notes the original Qwen3.8-Max previously held fourth place on the Frontend Code Arena, so this snapshot pushed it to first. Code Arena WebDev specifically evaluates real engineering front-end delivery, validating end-to-end capability beyond raw benchmark scores. The 0902 version continues strengthening coding and collaboration, while the sibling Qwen3.8-Flash remains positioned around lower training costs.

DataNorth reports that Alibaba released Qwen3.8-Max-0902 on QwenCloud on September 2, 2026, as a dated snapshot of its Qwen3.8-Max flagship model rather than a new model family. The snapshot targets complex software projects, long-running autonomous agents, and stronger native vision understanding, while preserving the The article documents developer-facing specifics for Qwen3.8-Max-0902: API ID qwen3.8-max-0902, with qwen3.8-max-2026-09-02 documented as an alias. The catalog entry lists a 1 million token context window, 128,000 token maximum output, a 256,000 token thinking budget, and support for function calling, built-in tools, a

Alibaba's Qwen3.8-Max-0902 snapshot took first place on Arena.ai's Code Arena: WebDev leaderboard on September 2, 2026 with 1,691 points, confirmed by TechNode, erasing a prior 37-point gap against Claude Opus 5 (Max) that had existed since the model's August 3 general availability. The update delivered the highest ben The 0902 suffix is a date stamp rather than a version increment: the 2.4-trillion-parameter MoE architecture, 95 billion active parameters, 1-million-token context window, and pricing are all retained from Qwen3.8-Max, and the QwenCloud model page confirms the alias qwen3.8-max-2026-09-02. The gains come from a targete

Gate News reported that Alibaba upgraded Qwen3.8-Max to the 0902 snapshot on September 2, with improved coding performance while retaining the 2.4 trillion parameter architecture and 1 million token context window. Across eight programming benchmarks every score increased, including TerminalBench 3.0 jumping from 11.3 to 29.0, DeepSWE 1.1 from 56.6 to 69.3, QwenSWEbench V2 from 55.1 to 70.0, and JobBench from 53.4 to 64.0. Compared to Claude Opus 5, Qwen3.8-Max-0902 surpassed the competitor on three benchmarks: MLS-Bench-Lite, SWE-Atlas QnA, and QwenSWEbench V2, while multimodal scores edged up by 0.4 to 3 points. The model is available via QwenCloud API at unchanged pricing of $2 per million input tokens and $6 per million output tokens.

Qwen's official X announcement on September 2, 2026 confirmed that Qwen3.8-Max was upgraded in place to a snapshot labeled Qwen3.8-Max-0902, with further post-training focused on coding and "cowork" workloads, and made it live through the QwenCloud API that same day. The snapshot retains the family's documented 2.4-tri However, Alibaba's public Model Studio page was updated on September 2, 2026 but still labels the model qwen3.8-max rather than exposing a separate qwen3.8-max-0902 SKU, so the dated suffix is not yet reflected as an independent billing row or benchmark suite. No matched 0902 benchmark, latency, or reliability results

Qwen3.8-Max-0902 (API alias qwen3.8-max-2026-09-02) went live on QwenCloud on September 1, 2026 at 10 PM ET as an in-place upgrade to the 2.4-trillion-parameter Qwen3.8-Max, further post-trained on "Coding & Cowork" while preserving the 1M-token context window, thinking mode, and tool ecosystem. Pricing is unchanged at Against Claude Opus 5, Qwen's published comparison table shows Opus 5 still ahead on TerminalBench, DeepSWE, NL2Repo, ProgramBench, SWE-Marathon, CoWorkBench, JobBench, and Toolathlon, while Qwen3.8-Max-0902 leads on repository code understanding, Automation Bench, two visual-reasoning benchmarks, and Qwen's in-house s

AI Release Tracker's page for Qwen3.8-Max-0902 confirms its release on September 2, 2026 by Qwen, seven days after Qwen3.8-Flash-Next. It documents the model as a 2.4 trillion parameter architecture with a 1 million token context window, listing benchmark results across QwenSWEBench V2, CoWorkBench, DeepSWE 1.1, NL2Rep Additional benchmark scores include DeepSWE 1.1 at 69.3% (7th of 26, best being Muse Spark 1.3 at 75.4%), NL2Repo-Bench at 64.9% (2nd of 6, best DeepSeek-V4.1-Flash at 65.4%), Terminal-Bench 3.0 at 29% (2nd of 7), CoWorkBench at 76.1% (best published), JobBench at 64% (2nd of 8), Toolathlon Verified at 73.3% (5th of 9)

DataCamp's blog provides a detailed update on Qwen3.8-Max-0902, released September 2, 2026 as a post-training upgrade to Qwen3.8-Max. The architecture remains unchanged at 2.4 trillion parameters with a 1 million token context, and pricing holds at $2 per million input and $6 per million output tokens. The update targe The article includes a full benchmark comparison table showing gains on TerminalBench 3.0 (11.3 to 29.0), ProgramBench Almost Solved (10.5 to 28.0), NL2Repo-Bench (55.9 to 64.9), QwenSWEBench V2 (55.1 to 70.0), and agent evaluations like WorkArena Elo (1348 to 1468) and CoWorkBench (74.8 to 76.1). Against Claude Opus 5

BenchmarkList aggregates benchmark scores specifically for Qwen3.8 Max 0902, listing JobBench at 64.0% (3rd of 36 models, 94th percentile), AutomationBench at 50.8% task success rate (4th of 42, 93rd percentile, flagged as the current field leader), and ClawEval-MM at 80.2% score (3rd of 11, 80th percentile). These res The page also compares Qwen3.8 Max 0902 against Claude Fable 5.1, Claude Opus 5, Claude Fable 5, GPT-5.6 Sol, Kimi K3, Qwen3.8-Flash-Next, and GLM 5.3 across benchmark categories including Agentic, Multimodal, and Reasoning evaluations. The aggregation provides per-eval rankings and percentile placements, giving techni

Kendr's research profile documents Qwen3.8 Max 0902 as an Alibaba Qwen updated snapshot with a 1M token context window, text, image, and video input modalities, and reasoning enabled by default. The profile lists prompt caching, reasoning controls, structured output, text generation, and tool calling among documented capabilities, with Alibaba Qwen identified as the model developer. According to the profile, the 2.4-trillion-parameter mixture-of-experts snapshot is post-trained for coding and agentic work, including multi-step software projects, multi-tool orchestration, and long-horizon task execution. It also targets chart reasoning, document parsing, and multimodal understanding over long documents and extended video, with configurable reasoning effort supported alongside tool calling and structured outputs.

Alibaba Cloud Model Studio issued a service-upgrade notice dated September 2, 2026, confirming the qwen3.8-max endpoint will automatically transition to the qwen3.8-max-0902 snapshot effective September 5, 2026 (UTC+8). The notice describes the upgrade as delivering breakthroughs in coding depth for complex engineering projects and long-horizon autonomous development. Billing items and pricing remain unchanged during the transition. The notice emphasizes significantly enhanced agentic collaboration capabilities for multi-tool orchestration and end-to-end delivery, alongside comprehensively refined visual understanding for chart reasoning and document parsing. The 1M context window, thinking mode, and full tool ecosystem are retained, and developers can test the snapshot early by specifying qwen3.8-max-0902 before the effective date via Model Studio.