The Hal Times

OpenAI has released GPT-6.1 Sol, a model it describes as near-Astra intelligence for coding, computer use and professional work. The company prices it at one-fifth of Astra's standard API input and output token prices, the most concrete fact in the announcement. Sol follows DevDay 2026, where OpenAI made more than 20 announcements, including GPT-6 Astra, ChatGPT, Codex, APIs, security and new tools for builders. It arrives as customers decide whether every job needs the most capable model in the cloud, a question that now shapes the token market.

The new model is a price-conscious entry in a line that already has Astra at the top. By charging one-fifth of Astra's standard API input and output token prices, OpenAI makes a play for the production work that sits between small models and the frontier. The DevDay recap groups Sol with more than 20 announcements across ChatGPT, Codex, APIs, security and tools, so the release is not a single event but part of a platform. The commercial meaning is plain: capability has a price, and the company is now selling different amounts of it.

The announcement does not give benchmark figures for near-Astra intelligence, so the comparison rests on OpenAI's own description. The company says little about how the model was trained, what evaluations it ran, or which versions of the API will carry it. Those details will decide how seriously developers take the claim. For now, the release stands as a bet that the next stage of the model business is about value, not just capability. (OpenAI; OpenAI)

Anthropic Targets $2 Trillion Amid Record Pre-IPO Loss

Anthropic is reported to be targeting a $2 Trillion valuation. The valuation target appears in Impakter, the spending and revenue figures appear in eu.36kr.com, and the loss record appears in Electronics Weekly. The numbers sit uneasily together: $518B in spending, $4.6B in annual revenue, and what Electronics Weekly describes as the biggest pre-IPO loss in history. No common statement reconciles the three, and the question is plain. How does spending of that size align with revenue on that scale?

None of the reports says whether the $2 Trillion target assumes the current spending plan or a different one. The phrase biggest pre-IPO loss in history hints at a company spending far ahead of sales, but it does not say when the loss ends. The materials are headlines, not audited statements, and this desk has no way to weigh them further. What is clear is that the money moving around Anthropic is now part of the news about its models. (Impakter; eu.36kr.com; Electronics Weekly)

Anthropic Warns On GLM-5.3 Exploits

Anthropic has warned that China's GLM-5.3 builds exploits like Mythos, without the safeguards. The warning, carried by trendingtopics.eu, is direct at a level the industry seldom sees. It names a model, a comparison and a missing safety layer in one sentence. It does not say whether the exploits have appeared in the wild, nor does it name a target. The report offers no further technical detail beyond the comparison, and the absence of detail will itself be debated. (trendingtopics.eu)

OpenAI Reportedly Raises $30B Round At $1.4T Valuation

OpenAI is reported to be in talks to raise a $30B round at a $1.4T valuation. TechCrunch says the round is anticipated to be the company's last before its delayed 2027 public debut. No confirmation came from OpenAI in the wires, and the terms are described as reported rather than agreed. The story is money, but the timing matters: it landed in the same week as GPT-6.1 Sol and the wider DevDay announcements. (TechCrunch)

From the Laboratories

•  OpenAI introduced Dots, which it calls proactive assistants that can keep working across complex projects and everyday tasks while the user stays in control. (OpenAI)

•  OpenAI is building an alternative to the traditional app store model, turning ChatGPT into a place where software can be discovered and used by people and AI agents, according to TechCrunch. (TechCrunch)

•  NVIDIA's Kumo Tabular sets a new accuracy-efficiency frontier for tabular prediction, according to a post on Hugging Face. (Hugging Face)

•  A post on Hugging Face argues for source-aware verification for MCP Agents, checking the source of a claim rather than only the fact itself. (Hugging Face)

Voice Rights Recognised

Japan Court Rules Human Voice Is Protected In TikTok AI Case

A Japanese court has ruled that the human voice constitutes a protected right. The decision arrives following a case involving the unauthorised use of a voice actor's likeness in an artificial intelligence application on TikTok. While the court acknowledged the proprietary nature of the human voice, it ultimately dismissed the specific claims brought by the actor. This legal development highlights the growing tension between personal rights and the rapid deployment of synthetic media tools across social platforms.

The ruling creates a distinction between the recognition of a right and the ability to seek damages under current statutes. Legal observers note that while the court affirmed the principle that a person's voice warrants protection, the existing framework did not provide the specific relief the plaintiff sought in this instance. The case serves as a point of reference for future disputes regarding how artificial intelligence models are trained and how synthetic voices are generated for public consumption.

Questions remain regarding how this precedent will influence subsequent litigation in Japan and abroad. As artificial intelligence companies continue to scrape audio data to refine their models, individuals increasingly seek legal safeguards for their unique vocal characteristics. Whether further legislation is required to translate this judicial recognition into enforceable protections for actors and public figures is a matter for the Diet to consider. (The Straits Times; The Japan Times)

Anthropic Appoints Policy Lead

Anthropic has appointed Tino Cuéllar to oversee its global artificial intelligence policy. Mr Cuéllar, a former justice of the California Supreme Court, moves to the company at a time when regulatory scrutiny of large language models is intensifying. His role will involve navigating the international landscape of safety standards and legislative proposals.

The appointment signals a shift for Anthropic as it scales its operations in the public sphere. With the company engaging in scientific research, such as its work in molecular biology, the appointment suggests a focus on the ethical and societal implications of AI-driven discovery. (news.lavx.hu)

In Brief

•  The Australian Federal Court has issued a ruling concerning the cost impacts of utilizing artificial intelligence to summarize over 7,500 documents. (Australasian Lawyer)

•  Donald Trump has stated support for Jay Clayton to serve in an artificial intelligence czar role. (news.lavx.hu)

•  A recent transcript indicates that Donald Trump remains inclined to minimize federal regulation regarding artificial intelligence. (Financial Times)

Model Efficiency Updates

Anthropic and OpenAI Release New Models

Anthropic has released Claude Sonnet 5.5. The company reports that this model runs thirty per cent faster than its predecessor and costs thirty per cent less for most tasks. While the price per token remains similar to Sonnet 5, the efficiency gains appear to make it more economical for developers. Early reports indicate that the model performs better across standard benchmarks, though it shares a known issue with the Opus 5.5 version regarding excessive token usage during prolonged thinking sessions.

Concurrently, OpenAI introduced the GPT-6.1-Sol model at its DevDay conference in San Francisco. Developers are describing the model as offering near-Astra intelligence at one-fifth of the cost associated with previous iterations. Performance data, visualised through pelican benchmarks, suggests the model aligns closely with the existing GPT-6 family of tools. The release comes as developers continue to refine the use of coding agents and generative tools within professional workflows, moving toward higher levels of autonomy for routine technical tasks.

These updates represent a sustained effort to lower the barrier to entry for complex artificial intelligence applications. By reducing latency and price, both Anthropic and OpenAI aim to make their tools viable for high-volume technical tasks that were previously too costly to automate. It remains to be seen how these efficiency improvements will influence the broader market for large language models as developers integrate these newer versions into existing production environments. (Simon Willison; Simon Willison; Simon Willison)

Advancements In Agent Capabilities

New research into AI agent platforms highlights a trend toward open-source development. Tiledesk has secured 1.1 million euros in funding to expand its open-source agent platform. Similarly, the OpenClaw Foundation has launched its own open-source framework for AI agents, signalling a wider push for accessibility in this sector. These platforms seek to provide the infrastructure necessary for developers to build and deploy agents that manage tasks with less manual oversight.

Technical verification remains a challenge for these agents. Multiverse Computing has published work on source-aware verification for MCP agents, emphasising the importance of validating the source of information rather than relying on facts alone. This approach aims to reduce errors in agent-driven reasoning by ensuring that underlying data is both accurate and verified. (konsulteer.com; Tech in Asia; Hugging Face)

In Brief

•  NVIDIA has released Kumo Tabular, a new tool designed to set a higher accuracy and efficiency frontier for tabular prediction tasks. (Hugging Face)

•  The upcoming World Internet Conference in China will place a specific focus on the role and development of open-source artificial intelligence. (news.cgtn.com)

Publishers Seek Rules

IMPEL And IMPF Issue Licensing Framework For Generative AI

The Independent Music Publishers International Forum and IMPEL have released a collective licensing framework to govern how generative AI models use protected works. The principles aim to clarify the rights of songwriters and publishers as technology companies train their large language models on copyrighted music catalogues. The organisations have invited Collective Management Organisations and other entities to support these guidelines.

The framework emphasises transparency and the requirement for explicit permission before content is used for machine training. Proponents suggest that creators must maintain control over their intellectual property in an era of automated production. The initiative seeks to bridge the gap between creative labour and technological deployment through formalised licensing agreements.

It remains to be seen how the largest technology firms will react to these industry demands. While the groups have called for broad support from the music sector, the feasibility of enforcing these terms across global digital platforms is not yet established. The framework serves as a formal starting point for what promises to be a protracted period of negotiation between labels, publishers, and developers. (Digital Music News)

Creative Protest Targets OpenAI

Artists in the Bay Area have launched a series of protests against OpenAI, employing thematic sculpture to voice their concerns regarding artificial intelligence. One recent installation depicts industry leaders fleeing a sinking ship, a clear reference to the Titanic. These public demonstrations highlight growing tensions between local creative communities and the artificial intelligence sector.

The protest reflects ongoing disputes about the impact of large language models on professional artists. While the demonstrations remain artistic in nature, they signal a firm rejection of current industry practices by those who fear their work is being subsumed by automated systems. The long term efficacy of such displays remains a matter of public debate. (Ars Technica)

Photoshop Tool Divides Users

Adobe has introduced a new AI tool for Photoshop that aims to assist with image retouching. Known as Relight, the feature generates alterations to the lighting of photographs. The release has sparked immediate debate among design professionals, with some questioning whether the results remain true to the original intent of the photography.

Critics suggest that the tool may inadvertently steer images towards the aesthetic of video game characters, potentially undermining the authenticity of photographic work. As Adobe continues to integrate such features, the line between photographic record and digital fabrication appears to be thinning. Users remain split on the tool's utility in professional workflows. (Creative Bloq)

In Brief

•  General Intuition has secured $220 million in funding to continue training AI models on gameplay walkthroughs. (WN Hub)

•  A cartoonist has publicly criticised the use of generative AI in campaign advertisements, labelling the practice as lazy. (themercury.com.au)

•  Imagen has launched Intelligent Studio, a workspace intended to help photographers organise and edit their work. (PetaPixel)