TechTimes, citing TechNode, reports that Qwen3.8-Max climbed to first place on Arena.ai's Code Arena: WebDev leaderboard on September 2, 2026 via the Qwen3.8-Max-0902 date-suffix snapshot — a reinforcement learning post-training pass rather than a new architecture, with the model retaining its 2.4-trillion-parameter Mo The article frames the upgrade within Alibaba's emerging convention of date-stamped post-training snapshots instead of version increments, referencing the August 26 Qwen3.8-Flash-Next open-weight preview as a parallel release pattern within the same wave. Beyond the benchmark wins, the piece notes that the underlying C

On September 1, 2026 (10pm ET), Alibaba's Qwen team released Qwen3.8-Max-0902 (alias qwen3.8-max-2026-09-02) on QwenCloud as an in-place post-training upgrade to the existing Qwen3.8-Max flagship, targeting coding and "cowork" agent work while keeping the 2.4T-parameter MoE base, 1M-token context window, thinking mode, According to the excerpt, on the benchmarks Qwen chose to publish, Claude Opus 5 still leads on TerminalBench, DeepSWE, NL2Repo, ProgramBench, SWE-Marathon, CoWorkBench, JobBench, and Toolathlon, while Qwen3.8-Max-0902 takes the lead on repository code understanding, Automation Bench, two visual-reasoning benchmarks, a

AireIter documents an important API-contract gap for Qwen3.8-Max-0902: although Qwen's September 2 announcement uses the dated suffix qwen3.8-max-0902, Alibaba Cloud Model Studio's documentation was updated September 2, 2026 but still labels the model simply as qwen3.8-max, with no separate qwen3.8-max-0902 row, benchm For developers this means Qwen3.8-Max-0902 should be treated as a live API revision of the qwen3.8-max family rather than a separately documented SKU, with no matched benchmark, latency, reliability, or token-efficiency results published for the 0902 snapshot specifically. The recommendation is to pin to whatever ident

DataCamp confirms that on September 2, 2026, Alibaba released Qwen3.8-Max-0902 as a post-training upgrade to Qwen3.8-Max with the same 2.4-trillion-parameter architecture, 1M-token context window, and unchanged $2 per million input / $6 per million output pricing, focused on coding and long-horizon agent work. All eigh The excerpt reports that Qwen3.8-Max-0902 now ranks first overall on Arena Code Arena WebDev with 1,691 points, three points above Claude Opus 5 Max (1,687) and 22 above the previous Qwen3.8-Max (1,669), and holds the top Pareto position at a blended $5 per million tokens. Against Claude Opus 5 it leads on three coding