Tech & AI News

The cost of running frontier AI just dropped off a cliff. On 30 July, OpenAI cut GPT-5.6 Luna's price by 80%, to $0.20 per million input tokens — a direct saving for high-volume API workloads, internal tools, and automation-heavy flows.

OpenAI's lineup now splits into clear tiers (Luna, Terra, Sol), so teams can send routine work to cheaper models and reserve deeper reasoning for harder jobs. For businesses building on AI in ASEAN, that means production-scale assistants and document pipelines are suddenly far more affordable.

ChatGPT crossed roughly 1 billion weekly active users this month, a milestone that puts generative AI squarely in the mainstream consumer and business stack. It is no longer a novelty — it is infrastructure people reach for daily.

The takeaway for companies: your customers and staff already use these tools. The question is no longer _whether_ to fold AI into your products and operations, but _how_ to do it reliably, privately, and close to your users.

Anthropic pushed coding and long-document work forward. Claude Fable 5 hit 80.3% on SWE-Bench Pro, one of the strongest coding benchmark scores reported this month — useful for cleaner automated pull requests and fewer manual reviews.

At the same time, Claude Opus 5 now supports a 1-million-token context window, enough to ingest full scripts or entire interview libraries in one pass. For legal, research, and strategy teams in the region, that means long-document analysis without chunking and reassembly.

Google leaned into always-on agents. Gemini 3.6 Flash cut output costs by 17% and offers up to 65% savings on long-horizon agentic tasks by using fewer reasoning steps, while Gemini Spark ($99.99/month, in the AI Ultra plan) runs as a cloud agent that keeps working even when your device is off.

Gemini also replaced Google Assistant system-wide on Android, a clear signal that AI is becoming the default layer in everyday workflows rather than a separate chat window.

Generation models moved well beyond text. Google Veo 3.1 sharpened video output for ads and training content; Lyria 3.5 added richer melodies, structured lyrics, and tempo controls for custom soundtracks; ByteDance Seedream 5.0 Pro brought region-precise image editing; and Adobe Research + Johns Hopkins "Wonder" turns a static image or video into a persistent 3D world you can navigate in six directions at 16 fps.

For ASEAN brands, that compresses the cost of localized creative — product demos, virtual walkthroughs, and campaign assets — from agency weeks to an afternoon.

As agents moved into real products, oversight caught up. LangChain's Deep Agents v0.7 cut input-token use by 65% through harness optimization, and Cekura ($30/month) added diagnosis and monitoring for autonomous agents in production — evidence the tooling around agents is maturing fast.

On the policy side, U.S. review gates, a deepfake law, and agent-disclosure pressure all added friction before launch. The simple planning takeaway for the rest of 2026: watch cost control, agent safety, and vendor choice.