Daily digest

13 items · ~13 min · Week 2026-W37

Must-read (1)

ARC Prize shows GPT-6 Astra's 99.9% ARC-AGI-3 score came from OpenAI's custom harness, not the model

OpenAI
Research official + media 4 src. ~1 min

ARC Prize's own writeup shows Astra scored 62.7% (for $26,098) on ARC-AGI-3 under its standard harness but 99.9% (for $18,817) under OpenAI's Provider Adapter, which preserves opaque reasoning state between requests; even with reasoning effort set to zero the adapter scores 96.7%. Separately, Fortune found five metrics in OpenAI's launch post were revised after publication (Astra's hallucination rate moved 4.2% -> 2% -> 4.2%; Fable 5.1's FrontierMath 87.8% -> 78% -> 83%), and ARC Prize disavows the AGI framing: 'we lack evidence to call this AGI yet.'

Why it matters
A benchmark organizer formally documenting that harness choice, not model capability, produced the headline AGI-claim score — a template case for benchmark transparency.

Worth knowing (5)

Zhipu's first interim report as a listed company: revenue up 400%, API now 86.5% of the business

Zhipu AI (Z.ai)
Industry media only 3 src. ~1 min

Zhipu's (02513.HK) first half-year report since its IPO, widely covered on Sept 6-7, shows H1 2026 revenue of RMB 954 million, up 399.7% year over year and already above full-year 2025. Open-platform/API revenue grew 2735.7% to RMB 825 million and now makes up 86.5% of revenue (15.2% a year ago), while net loss narrowed 12.1% to RMB 2.07 billion; ARR hit $1.6 billion by end of August.

Why it matters
Proof that a Chinese open-weights lab can flip from on-prem deployment deals to API subscription revenue at scale, with API gross margin turning positive at 24.6%.

OpenAI says its 'automated research intern' goal is met, targets fully automated researcher by March 2028

OpenAI
Research official + media 3 src. ~1 min

In a Sep 6, 2026 blog post ('Research acceleration: The view inside OpenAI') OpenAI says it reached the goal set in fall 2025 of an 'automated research intern' — a system handling well-defined tasks that would take a skilled researcher days — and is on track for an automated AI researcher by March 2028. Internal figures: 3.1 agent-workdays per human workday as of mid-August 2026, median researcher spending over $600/day on inference and the 90th percentile over $7,000/day.

Why it matters
First quantified inside look at how much frontier-lab research work agents already do, with a dated milestone toward automated AI research.

OpenAI chief scientist warns in 'An Alien Mind' essay that no one is prepared for recursive self-improvement

OpenAI
Research official + media 2 src. ~1 min

OpenAI's chief scientist published a Sep 6, 2026 essay arguing internal results make him 'strongly' expect current progress speed to continue into recursive self-improvement, writing 'I am concerned no one is prepared for the consequences of a continued rapid rise in machine intelligence.' He says OpenAI will keep building defenses and may 'unilaterally withhold further scaling as needed,' but argues broader interventions beyond lab self-restraint are required.

Why it matters
The most senior safety voice at a frontier lab publicly calling for external intervention, not just voluntary lab caution.

RoboTok: internet-scale retrieval of human demonstrations for dexterous robot policies

Research official 2 src. ~1 min

RoboTok is a data engine that, given a query manipulation video, retrieves relevant human demonstrations from web video for training dexterous robot policies. It learns a latent motion space from 3D hand trajectories in actor-centered frames, making retrieval robust to camera viewpoint, scene appearance and occlusion while remaining searchable at internet scale.

Why it matters
Over 100 upvotes on HF Daily Papers (115) — targets the core bottleneck of robot learning: cheap, diverse, long-tail training data.

LatentPress: compressing context into continuous memory tokens a frozen LLM reads directly

Research official 2 src. ~1 min

A two-author paper introduces a third context representation beyond text and screenshots: continuous memory tokens written into a frozen decoder through its input-embedding interface, with no text reconstruction at inference. A small reader-matched writer (4.2M-26.2M adapter parameters, about 0.1% of the decoder) compresses 4-16x; on LongMemEval it scores 0.504 at 7.7x compression versus 0.490 for uncompressed evidence, far above text summaries (0.184) and OCR-based compression.

Why it matters
Over 100 upvotes on HF Daily Papers (110) — shows latent token memory can beat textual summarization for long-context recall at near-zero training cost.
For reference (7)

Morgan Stanley briefing: MiniMax ARR at $800M and Zhipu at $1.6B as model competition turns into a tiered elimination round

MiniMax / Zhipu AI
Industry media only 2 src. ~1 min

A Morgan Stanley private briefing reported on Sept 6-7 puts MiniMax's end-August ARR at $800M (weekly basis) with a raised analyst view of $1.3B by year-end, and Zhipu's ARR at $1.6B in August ($2B+ on a weekly basis) with year-end guidance lifted to $2.4B. The same briefing dates GLM-6 for October on GLM-5.3-Flash's architecture and lists MiniMax M3.1, H3.1 and M3 Pro for H2 plus a 10T-parameter model planned next year; MiniMax shares rose nearly 8% on the news.

Why it matters
First concrete roadmap detail on GLM-6 and MiniMax's next-gen models, and a frame for why pure price competition stops working below SOTA capability.

Alibaba's Qwen Office passes 30 million users within about a month, more than half from enterprises

Alibaba (Qwen)
Industry media only 3 src. ~1 min

Coverage on Sept 4-7 reports Alibaba's Qwen Office (千问办公) agent reached 30 million users within roughly a month of launch, with over 52% of accounts tied to real enterprise organizations, driven by DingTalk distribution rather than ad spend. The product introduced paid subscription tiers on August 10 — the second major Chinese AI platform to do so after ByteDance's Doubao — and positions Qwen3.8 as the model layer across DingTalk, Alibaba Cloud and CRM/ERP/OA systems.

Why it matters
The fastest enterprise-agent ramp in China so far, showing Alibaba's model-plus-workplace-ecosystem bundling as a distinct go-to-market against Tencent and ByteDance office agents.

GPT-6 Astra reaches the $20 ChatGPT Plus tier after a rollout Altman had to apologize for

OpenAI
Models / LLM official + media 4 src. ~1 min

Following up on the Sep 3 flagship launch: on Sep 6, 2026 OpenAI began rolling GPT-6 Astra out to $20 ChatGPT Plus subscribers; an OpenAI X post confirmed availability 'to all Pro, Enterprise, and Business Premium users in ChatGPT Work and Codex. It's also live in the API.' The rollout to paying users was delayed days after launch, prompting a public apology from Sam Altman for the 'messy' release; free users still only have GPT-5.6 Sol and GPT-5.

Why it matters
Completes consumer availability of OpenAI's flagship model and shows deployment, not training, is now the bottleneck for frontier releases.

Editable Visual Design: coding agents generate layered HTML/CSS visuals instead of flattened images

Research official 2 src. ~1 min

A team around Junyan Ye proposes a new visual-generation paradigm where a coding agent uses a VLM as the creative brain and an image model as an on-demand visual-world simulator: the agent generates isolated assets, writes native HTML/CSS, and refines the design against rendered feedback. The result is editable, layer-decoupled artifacts, in contrast to end-to-end diffusion output (GPT-Image-2, Nano Banana) which yields flattened bitmaps with error-prone text.

Why it matters
Top-upvoted paper of the Sep 5 HF chart (42 upvotes); points at a post-image-editor direction where designs stay structured and editable rather than baked into pixels.

Last Translation Benchmark: human-authored tasks that still break frontier translation models

Research official 2 src. ~1 min

A large consortium (Vilem Zouhar, Niyati Bafna, Stella Biderman and others) releases LTBv1, a live collection of human-authored, peer-reviewed texts, images, audio and video that break leading machine translation models as standard benchmarks saturate. Each example ships with handcrafted verification rules describing concrete failure modes, enabling reliable, reward-hack-resistant, actionable evaluation.

Why it matters
An unsaturated, community-maintained eval with rule-based verification — a template for post-saturation benchmarking beyond MT.

RealSWE: coding-agent benchmark rebuilt around real, casual user requests

Research official 2 src. ~1 min

The authors quantify the gap between SWE-bench-style tasks and real usage: 88% of real prompts carry only a bare problem statement (versus 7% of benchmark problems), and 87% are casually written versus 94% formal in benchmarks. RealSWE derives 381 multi-variant task families from SWE-bench Verified and Pro to test compositional behavior under realistic request styles.

Why it matters
Argues current coding-agent scores overstate real-world readiness by measuring on curated, information-rich issues rather than how users actually write.

Krea connects to ChatGPT for image and video generation inside conversations

Krea
Tools official 1 src. ~1 min

On September 5, 2026, Krea announced a ChatGPT integration that lets users generate images and video with Krea's model lineup directly inside a ChatGPT conversation. Existing Krea moodboards and styles carry over into the chat workflow, and outputs can be saved back to the user's Krea library.

Why it matters
Another step in ChatGPT becoming a front end for third-party generative media models, with Krea's style/moodboard system as the differentiator.