GPT-6.2 Astra Launches, Anthropic IPO Prices at $83, Qwen4-Max Debuts
OpenAI's GPT-6.2 Astra swept ARC-AGI-3 with a perfect 100% on the official validation test, ranking second on the Artificial Analysis Intelligence Index behind only Claude Fable 5.2, with API input and output tokens priced 20% below GPT-6 Astra. Its system card also marks the first time OpenAI has rated a model High on self-improvement. Days later, Anthropic priced its Nasdaq IPO at $83 per share, giving regional investors a reference point for AI valuations. On the Chinese side, Alibaba's Qwen4 Max scored 43 on the Intelligence Index at $3.27 per task, while Tencent open-sourced the 770B-parameter multimodal MoE Hunyuan Hy4 with a 1M context window. For Singapore and ASEAN builders, the week delivered more flagship options and cheaper inference.
Five things this week
Enough to retellUS vs China
The same week, both sides| Dimension | United States | China | What it means for ASEAN |
|---|---|---|---|
| New models | OpenAI ships GPT-6.2 Astra and Sol; Astra scores 100% on ARC-AGI-3 validation | Alibaba launches Qwen4-Max; Tencent open-sources 770B multimodal MoE Hunyuan Hy4 | More flagship options and usable open-weight alternatives for self-hosting |
| Prices | GPT-6.2 Astra API input/output tokens 20% cheaper than GPT-6 Astra | Qwen4 Max costs $3.27 per task, positioned as value play | Falling inference costs make models affordable for ASEAN small teams |
| Benchmarks | Astra ranks second on Intelligence Index, behind Claude Fable 5.2, five points above GPT-6 Astra | Qwen4 Max scores 43, up nine from Qwen3.8 Max, trailing Kimi K3.1 Pro | US-China gap narrows; choose models by value, not origin |
| Policy | GPT-6.2 Astra system card rates self-improvement High, cybersecurity Critical | — | Watch safety ratings for enterprise compliance and risk assessments |