---

I. 3 Key Updates

Key Update 1 | OpenAI Suddenly Cuts Prices — GPT-5.6 Luna Drops by 80%, Global Token Price War Intensifies

On July 30 local time, OpenAI announced price cuts for two models in the GPT-5.6 series: The lightweight model Luna sees input prices drop from $1 to $0.2 and output prices from $6 to $1.2 (per million tokens, an 80% reduction); the mid-tier Terra drops by 20% to $2/$12; the flagship Sol remains unchanged in price but introduces a new Fast mode (up to 2.5x speed at 2x price), replacing the original Priority Processing. These price cuts are simultaneously reflected in the quota consumption of Codex and ChatGPT Work—subscription prices remain unchanged, but the same quota allows the model to do more work. This comes just three weeks after the release of the GPT-5.6 series on July 9. OpenAI stated directly: Cheaper Chinese open-source models (such as Kimi K3, DeepSeek V4) are exerting pressure. Third-party calculations show that at equivalent intelligence levels, Luna High's cost-performance ratio is still slightly inferior to DeepSeek V4 Pro.

LLM Expert | The price cuts target volume-driven tiers rather than flagships. Essentially, this is token deflation driven by dual factors: "efficiency improvements + pressure from Chinese open-source." Capability ceilings continue to command high prices, while volume capabilities are being pushed to floor prices. OpenAI is selling the models that support Agents doing work at the price point of past simple small models. Value distribution is shifting from "selling models" to "using models to produce results."

Software Expert | Three groups benefit most directly: ① API users—no changes needed, bills automatically shrink by 80%; ② ChatGPT/Codex subscription users—quotas remain unchanged but Luna/Terra consume less, high-frequency tasks can now be used freely; ③ teams currently selecting vendors—include Luna in head-to-head testing with DeepSeek V4 before deciding. Action Item: Recalculate your AI cost sheet this week.

Key Update 2 | ByteDance Major Restructuring — Feishu Merges into Doubao, Ending the "Horse Race Era" of AI Office

On the morning of July 30, ByteDance internal memo announced: The Feishu product team and Doubao product team have been integrated to form a new Doubao product team, overseen by Doubao lead Zhao Qi, with Feishu CEO Xie Xin reporting to Zhao Qi; Feishu sales/marketing teams merged with Volcano Engine to form a ToB department called "Creativity Service Platform," led by Volcano President Tan Dai. Feishu has sunk from an independent business group to a capability component; existing product services remain unchanged; the Doubao Enterprise Edition co-built by the Feishu team is already in beta testing with some Feishu customers. This is the third major tech giant integrating its AI office line within ten days: Tencent integrated QClaw into WorkBuddy on July 20, Alibaba integrated three Agent products into Qianwen Office on July 27, and ByteDance merged Feishu into Doubao on July 30.

LLM Expert | ByteDance's logic is clear: AI is now the sole entry point for enterprise procurement. Feishu's documents/meetings/group chats are "infrastructure," while Doubao is the "brain." The independent narrative of collaborative office tools is disappearing, forming a top-three landscape: Tencent WorkBuddy (June monthly visits 20.97 million, leading in entry points) vs. Alibaba Qianwen Office (DingTalk + Tongyi) vs. ByteDance Doubao (Feishu + Volcano Engine).

Software Expert | Feishu users: Zero short-term impact, no action required; medium-to-long term, watch for Doubao AI capabilities accelerating into documents and meetings—for enterprises, this is a free upgrade. Enterprises currently selecting office AI: Stop single-point selection, choosing based on ecosystem alignment (WeChat system/DingTalk system/Feishu system) is more practical than comparing features individually.

Key Update 3 | Tencent WorkBuddy V5.3.5 Launches "Human-AI Dual Writing" — Same Document, Humans and AI Write Together

On July 30, Tencent WorkBuddy released V5.3.5, launching "Human-AI Dual Writing" collaborative editing in partnership with Tencent Docs: Users can co-create and modify with AI within the same document, supporting mainstream formats like Word, Excel, PPT, and Markdown, compatible with local Office files and Tencent Docs online documents; it supports selecting specific areas for AI to locally modify while keeping the rest unchanged; AI co-created content can be one-click uploaded to Tencent Docs, enabling real-time multi-user, human-machine collaboration across devices.

LLM Expert | The value of "Human-AI Dual Writing" lies in the interaction paradigm: shifting from turn-based "issue command-wait for result" to real-time co-creation on the same canvas. This requires extremely low latency and precise local editing capabilities (in-context editing) from the model, marking a key leap for office AI from toy to tool.

Software Expert | Heavy document users writing proposals or revising reports: Update today and try "select area for local modification". This addresses the highest frequency daily pain point—the era where asking AI to change one paragraph resulted in rewriting the whole text is over. Team collaboration users remember to one-click upload co-created docs to Tencent Docs so versions don't get messy.

II. 10 Curated Updates

1. Musk: Grok 4.6 to Release in One Week (July 30): 1.5 trillion parameters, significant improvements in SFT/RL; Grok 4.7 (2.1 trillion) to follow in weeks. → X users can touch it next week; the parameter race isn't over.

2. Moonshot AI Completes Over $3.5 Billion Series F Funding (Reported July 30): Valuation at $35 billion, pushing for HK Pre-IPO as early as 6 months. → Kimi has ample ammunition; investment in the K3 ecosystem will only increase.

3. Alibaba Qianwen Office Integrates Three Agent Products (Supplement · July 27): Connecting DingTalk + Tongyi. → End of the three-way horse race; DingTalk users await feature integration.

4. Tencent WorkBuddy Integrates QClaw (Supplement · July 20): Code agent merged into office entry point. → WorkBuddy expands from office to R&D scenarios.

5. Microsoft Records Largest Single-Day Gain in Nearly 18 Years (US Stocks July 30): Driven by AI business. → Capital markets continue voting for "AI monetization."

6. China's AI Token Call Volume Ranks First Globally for 13 Consecutive Weeks (Statistics July 30). → China has substantively led in application layer activity.

7. Over 1,000 AI Employees Including OpenAI Staff Sign Joint Petition (July 30): Calling for control over AI development speed. → Industry self-regulation voices rising; watch subsequent regulatory trends.

8. Doubao Enterprise Edition Opens Beta Testing: Natively integrates Feishu docs, sheets, meetings, group chats, with enterprise data isolation and auditing. → Feishu customers can apply to try it out.

9. Altman: Over-reliance on AI May Lead to Decline in Human Cognitive Abilities (July 30). → It's the right time to set an "AI usage posture" for your team.

10. Apple FY2026 Q3 Revenue $109.4 Billion (Early Morning July 31): YoY +16.4%, AI features driving device upgrades. → The monetization path for edge-side AI is materializing.

III. News Flash

• OpenAI replaced the Auto-review model for ChatGPT and Codex CLI from GPT-5.4 to GPT-5.6 Luna.

• New GPT-5.6 pricing effective July 30, subsequently synced to AWS platform.

• Analysys Q2: Tencent WorkBuddy June monthly visits 20.97 million, #1 in domestic office AI.

• Doubao App MAU 382 million (QuestMobile); Doubao large model daily token calls 180 trillion.

• Over 90% of Feishu's new clients in Q2 2026 also purchased AI products.

• Xueqiu Comment: Drop in token unit price benefits high-token-consumption Work/Agent applications (Jevons Paradox).

IV. Tomorrow's Focus

1. Grok 4.6: Musk says release in one week (approx. Aug 7);

2. WAIC 2026: World Artificial Intelligence Conference ongoing, watch for new releases at major vendor booths;

3. MiniMax & ByteDance Video Model Moves: Rumors dense, new releases possible anytime;

4. Zhipu GLM Coding Plan: Developer subscription expansion updates.

---