Qwen3.8-Max-0902
Vintage: 2026-09. Primary evidence is official @Alibaba_Qwen X (2026-09-02/03) plus independent @arena board-operator posts the same night, as hydrated in 2026-09-03-x-ai-pass-3-sep-2026-open-model-fleets-agentic-coding-price (
fetch_method: x-mcp). This is a closed API SKU on QwenCloud, not an open-weight drop. Carry the Code Arena 3-point lead as a narrow / statistical tie at the top, not categorical supremacy.
Qwen3.8-Max-0902
One-line summary: Alibaba Qwen's 2 Sep 2026 closed API upgrade (2.4T parameters, 1M context, $2/$6 per 1M tokens) that sits statistically tied at the top of Code Arena WebDev and, per the arena operator, at the top of that board's price/performance Pareto frontier.
What it is
A point update of Qwen3.8-Max, announced by @Alibaba_Qwen: "2.4T parameters. 1M context tokens," further post-trained on coding and "Cowork," live via API on QwenCloud. Closed API, not open weights. Do not file this as a chinese-open-weight-frontier-parity crossing.
Why it matters to this thread
Frontier pricing and agentic-coding evaluations are in-scope. The durable fact is the price axis — blended $5/MToken Pareto-frontier placement, and $2/$6 headline vs claude-fable-5-1's $10/$50 — not a 3-point WebDev margin. Arena.ai corroborating its own board is consistency, not a second measurement.
Key facts (from 2026-09-03-x-ai-pass-3-sep-2026-open-model-fleets-agentic-coding-price)
- From the same source (@Alibaba_Qwen, 2026-09-02): 2.4T parameters; 1M context; $2 / MTok input and $6 / MTok output; cache hits $0.17 (explicit) and $0.25 (implicit); live on QwenCloud.
- From the same source (@Alibaba_Qwen): #1 on Code Arena WebDev, "jumps from 1669 to 1691."
- From the same source (@Alibaba_Qwen): #1 overall on Code Arena and "top of the Pareto frontier at $5/MToken."
- From the same source (@arena): board operator — debuted at #1 overall in Code Arena: WebDev with 1,691 points, 3 pts above Claude Opus 5 (Max), 17 pts above Kimi K3 (Max), 22 pts above the previous Qwen3.8-Max. Carry as statistically tied at the top — three points, not thirty, is inside the noise band of most arena boards.
- From the same source (@arena): at a blended $5/MToken it became "the highest-scoring model on the Pareto frontier," sitting above HY4 Preview (1,629 pts at $2.08/MToken).
- From the same source (@Alibaba_Qwen, 2026-09-03): introduced E-Commerce Bench — agents start with ¥100,000 and run online stores for 365 simulated days (sourcing, negotiation, pricing, promotions, inventory, cash flow; "real e-commerce data"). Vendor-authored; treat as a claim about what should be measured, not an independent result. See agi-definitions-and-benchmark-saturation / ai-coding-benchmarks.
What this source does not establish
- Not an open-weight release. Do not attach as a Kimi/GLM-style weights drop.
- Not "beats Claude Opus 5 (Max)." The 3-point margin is a narrow / statistical tie. Board methodology and vote counts matter at that gap.
- Arena.ai is the board operator. Its posts corroborate Qwen's claim about Arena.ai's own board — consistency, not independent replication.
- HY4 Preview / Claude Opus 5 (Max) are comparison chips only. No entity pages minted for them.
- E-Commerce Bench scores are not in this pass. The post introduces the bench; it does not report a winner table here.
- Launch-page / model-card URLs were not fetched. Specs and prices are as written on X.
- Sep 3 Astra attach is price-class color only. From 2026-09-03-gpt-6-astra-rolls-out-computer-use-flagship: gpt-6-astra standard short is $10 / $50. This page's already-filed $2 / $6 remains a different price class. Not a Qwen rewrite.
Open questions
- Does the 3-point WebDev lead survive a later vote dump or a methodology note from Arena.ai?
- Is the blended $5/MToken Pareto figure a published rate card or an arena-side blend of the $2/$6 list?
- Who tops E-Commerce Bench, and is that result independent of the vendor that wrote the bench?
Sources
- 2026-09-03-x-ai-pass-3-sep-2026-open-model-fleets-agentic-coding-price
- 2026-09-03-gpt-6-astra-rolls-out-computer-use-flagship — adjacent $10/$50 vs $2/$6 price-class color only