GPT-5.6 launch nears; Fable 5 blocked by US export order
GPT-5.6 targets this week with 1.5M-token context; Fable 5 sidelined by export controls; Trump signs AI EO; Meta cuts 8,000 jobs.
By BINA Editorial
OpenAI's next flagship model is within days of release, a US export control directive has frozen access to two frontier models globally, the White House has drawn a policy line against mandatory AI licensing, Meta is reshaping its workforce around artificial intelligence at a cost of 15,000 positions, and an open-source challenger has cleared the GPT-5.5 bar on code-engineering benchmarks.
GPT-5.6 opens its launch window with 1.5 million tokens and alignment improvements
OpenAI is tracking a release for GPT-5.6 inside a June 22–28 window, with prediction-market platform Polymarket assigning 83 percent probability and more than $960,000 in open bets to that range. OpenAI Chief Scientist Jakub Pachocki described the model to staff as "a meaningful improvement" over GPT-5.5 — the first named-executive statement to reach the public, shifting coverage from developer speculation to something closer to an unofficial preview. The headline additions are a context window expanded to 1.5 million tokens, 43 percent above GPT-5.5, and a targeted fix for the alignment regressions that accompanied GPT-5.5's April launch. The model is also expected to deliver 10–15 percent further gains in token efficiency, continuing the roughly six-week flagship cadence OpenAI has maintained since the start of the year. No official announcement or system card has been published.
US export controls sideline Fable 5 globally as its free-trial window closes
On 12 June the US Department of Commerce issued an export control directive blocking global access to Fable 5 and Mythos 5. The timing is particularly disruptive for subscribers: Fable 5 was also scheduled to move from a free-trial model to usage-credit billing on 23 June, meaning users lose free access and paid access on the same day. The models had attracted significant developer adoption since their 2026 release for their reasoning capabilities and extended context. No timeline for resolution has been published and no public statement has addressed the scope of the controls or a projected lifting date. The move follows an earlier June 12 directive and appears to be part of a broader series of technology export restrictions targeting frontier AI systems.
Trump signs executive order creating a voluntary framework for frontier AI
The White House released an executive order titled "Promoting Advanced Artificial Intelligence Innovation and Security" on 2 June, establishing what the administration described as two parallel tracks: strengthening federal and private-sector cyber defences against AI-enabled attacks, and building a voluntary benchmarking and review framework for frontier models. A centrepiece is an AI cybersecurity clearinghouse intended to help government and industry identify and remediate software vulnerabilities at scale. The order explicitly prohibits interpreting its provisions as authorising any mandatory licensing, pre-clearance, or permitting regime for developing or distributing AI models — a direct signal to the industry that the administration does not intend to impose the pre-release approval requirements some lawmakers have proposed. The White House fact sheet notes that the Attorney General is separately directed to prioritise enforcement against individuals who use AI to carry out computer intrusions, data theft, or other criminal activity.
Meta restructures around AI, cutting 8,000 jobs and reassigning 7,000 to new teams
Meta began notifying approximately 8,000 employees of redundancies in what Chief People Officer Janelle Gale framed as a necessary step to fund the company's AI buildout — the largest single round of cuts since CEO Mark Zuckerberg's 2022–23 "Year of Efficiency" campaign removed roughly 21,000 positions. Beyond the layoffs, roughly 7,000 additional workers are being redirected into newly created AI-focused units with names such as Applied AI Engineering and Agent Transformation Accelerator, bringing the total workforce shift to around 15,000 people — close to one-fifth of the company. Meta's projected capital expenditure for 2026 runs between $125 billion and $145 billion, more than twice its 2025 outlay, reflecting the infrastructure scale the company believes is necessary to compete across AI services, recommendation systems, and the metaverse.
GLM-5.2 surpasses GPT-5.5 on code benchmarks under an open MIT licence
Zhipu AI released GLM-5.2 on 13 June under a permissive MIT licence, making it one of the most capable openly licensed large language models currently available for local deployment. The model features a usable 1 million-token context window and scores 62.1 on the SWE-bench Pro software-engineering benchmark, outperforming GPT-5.5's 58.6 and placing GLM-5.2 above several closed-source systems on that task. The release gives developers a high-capacity option for on-premises use without API usage costs or the data-handling constraints of closed commercial models. Concurrently, the IplanRIO organisation published Rio 3.5 Open 397B on Hugging Face under an MIT licence, another signal that the upper end of the open-weight capability range is narrowing the gap with leading proprietary systems.