Alibaba launched a preview of its flagship Qwen3.8 Max model on Sunday, describing it as second only to Anthropic's Fable 5 among frontier AI systems — a striking claim for a model released with virtually no public technical benchmarks.
The Hangzhou-based company announced the release in a post on X, noting that Qwen3.8 Max carries 2.4 trillion parameters. Beyond that figure, Alibaba has disclosed little else: no benchmark scores, no context window specs, no architecture details beyond the parameter count. The company said an open-weight release is planned but has not given a timeline.
A Benchmark Claim Without the Benchmarks
Positioning a model as "second only to Anthropic's Fable 5" is an unusually direct competitive statement — particularly for a preview release. Alibaba hasn't published the evaluation results that would support that ranking, and the claim sits in tension with how AI labs typically validate frontier-tier positioning: through third-party evals, leaderboard submissions, or detailed technical reports.
That doesn't make the claim implausible. Alibaba's Qwen team has a strong track record, and the Qwen3.8 architecture was already live in preview before Sunday's broader announcement. But until independent benchmarks arrive, the frontier-tier positioning remains self-reported.
For context, Moonshot AI's Kimi K3 — which carries 2.8 trillion parameters — topped LMArena's Code Arena leaderboard at launch with third-party results to show for it. Qwen3.8 Max hasn't yet cleared that bar publicly.
The Kimi K3 Shadow
The timing of the release is hard to separate from competitive dynamics. Moonshot AI unveiled Kimi K3 earlier in the week, triggering a wave of developer excitement, a 500% surge in Google Trends interest, and a selloff across global AI-adjacent stocks. The Chinese startup's open-weight model became a benchmark story that upended Western AI market perceptions within days of launch.
Alibaba's Sunday release — a preview drop on a model with more parameters than Kimi K3 — reads, in part, as a signal that the Qwen team isn't ceding the frontier open-weight space. Qwen3.8 Max's 2.4 trillion parameters is large, though Kimi K3's 2.8 trillion gives Moonshot a numerical edge on raw scale. Whether scale translates to capability will only be settled once independent evaluations run against both models.
Where the Model Is Available — and at What Price
Developers can access Qwen3.8 Max today through Alibaba's Qoder platform, which covers the desktop client, JetBrains plugin, CLI, mobile app, and cloud agents. To accompany the launch, Qoder is running a time-limited pricing promotion: the model's Credits billing coefficient is reduced from the standard 0.5x to 0.05x during regular hours — effectively a 90% discount, or 10x the usage for the same cost.
During off-peak hours (22:00–08:00 Singapore Time), that discount deepens to 98% off, reducing the coefficient to 0.01x. The promotion applies automatically for Pro Trial and paid subscribers after updating to the latest Qoder version; no sign-up is required. The end date has not been announced.
The discount structure is notable not just for its depth but for what it signals: Alibaba is willing to absorb significant compute cost to drive adoption of a flagship preview model. That's a familiar playbook — Chinese AI providers have increasingly used aggressive access pricing to build developer mindshare ahead of a model's full release.
The Apple Intelligence Angle
Alibaba's frontier ambitions are also playing out in an adjacent arena. Beijing recently approved Apple Intelligence for use in China, with Alibaba serving as the technology partner powering Apple's AI features for Chinese users. That partnership gives Alibaba distribution at iPhone scale inside the world's largest smartphone market — a strategic asset that's independent of how Qwen3.8 Max ultimately benchmarks.
BABA shares fell 2.14% on Sunday despite — or perhaps because of — the preview launch. Markets may be reacting to the absence of hard technical data, or simply digesting the broader week of volatility triggered by Kimi K3's debut.
What to Watch
The open-weight release will be the real test. When Alibaba publishes model weights alongside reproducible benchmark results, the "second only to Fable 5" claim becomes verifiable — or it doesn't. Until then, Qwen3.8 Max occupies an unusual position: a self-declared frontier model with strong institutional credibility behind it, but no public evidence trail yet.
Developers who want to stress-test the model ahead of that release can do so now through Qoder at a fraction of normal cost. The 98% off night-time rate, in particular, makes extended experimentation cheap enough that the lack of published benchmarks may not slow adoption much — especially for teams already building on the Qwen ecosystem.
Related Reading
- Qwen3.8's 2.4T Parameters Are Already Live — and Open Weights Are Coming
- Kimi K3 Is Official: The 2.8T Open Model That Built Its Own Compiler and Designed Its Own Chip
- Moonshot AI Searches Explode 500% — and the Markets Already Know What That Means
- 10 Million Kimi K3 Tokens Are Free Tonight — Here's How to Claim Your API Key
