AI News: April 26, 2026

1. Alibaba’s Qwen3.6-27B Beats the 397B Predecessor on Coding and Closes In on Claude Opus 4.5

Alibaba. Alibaba released Qwen3.6-27B, a 27-billion-parameter dense model that outperforms its much larger 397B Qwen predecessor on most coding tasks while shipping as open weights on Hugging Face and ModelScope. Reported scores include 77.2 on SWE-bench Verified (up from 76.2) and 59.3 on Terminal-Bench 2.0 (up from 52.5), with multimodal and reasoning numbers on GPQA Diamond and MMMU close to Claude 4.5 Opus. The combination — dense rather than MoE, broadly capable, openly available — makes Qwen3.6-27B one of the more attractive bases for self-hosted coding agents announced this month. Source

2. Federal Reserve Study Estimates 500,000 Fewer US Programmer Jobs Since ChatGPT

Federal Reserve. A new Federal Reserve paper estimates that US programmer employment is roughly 500,000 jobs below the counterfactual trend that would have held without ChatGPT, with growth in programming-heavy occupations dropping from nearly 5% per year to essentially flat. The methodology cross-references monthly household-survey employment data with Department of Labor occupational classifications, isolating the AI-attributable share at about 3 percentage points of annual decline since late 2022. Notably, wages have stayed stable — the gap shows up entirely in hiring volume — and the effect did not appear until mid-2024, suggesting employers needed time to evaluate AI capability before adjusting headcount plans rather than reacting to launch hype. Source

3. Maine Governor Vetoes First-in-Nation Data Center Moratorium

Maine. Governor Janet Mills vetoed L.D. 307, which would have imposed the first statewide US moratorium on new data center permits through November 2027, citing the bill’s lack of an exemption for a supported project in Jay even while acknowledging the underlying environmental concerns. The veto matters beyond Maine: with Democratic-led states wrestling with whether to slow down hyperscaler-funded data center buildouts, Mills declining to sign signals that even sympathetic governments will struggle to enact blanket pauses on AI infrastructure. Expect similar bills with project-specific carve-outs to surface in other states this year. Source

4. UAE Plans to Run Half of Its Government Through Autonomous AI Agents Within Two Years

UAE. The United Arab Emirates announced a target to operate 50% of government sectors, services, and processes via autonomous AI agents within two years, framed as the first national-scale push to put autonomous systems behind core public-sector workloads. All federal employees will receive training to work alongside the agents, which the UAE positions as “executive partners” rather than replacements for staff. Announcement-light on partnerships and procurement details so far, the plan is best read as a firm political commitment that will pull in vendors competing for sovereign-AI mandates over the rest of 2026. Source