Fresh stories
OpenAI says GPT-5.6 Sol launches publicly Thursday
OpenAI says GPT-5.6 Sol, Terra, and Luna will launch publicly Thursday as preview access expands. Testers describe Sol as strong for coding, agents, and computer use; Wafer reports Cerebras serving up to 750 tokens/sec.
Meta launches Muse Image with reasoning, search, code, and self-refinement
Meta launched Muse Image in its apps and previewed Muse Video from the same media-generation family. Meta says Muse Image can reason, search, write code, self-refine, and use test-time compute before generating images.
OpenAI says GPT-5.6 Sol launches publicly Thursday
OpenAI says GPT-5.6 Sol, Terra, and Luna will launch publicly Thursday as preview access expands. Testers describe Sol as strong for coding, agents, and computer use; Wafer reports Cerebras serving up to 750 tokens/sec.
Claude Cowork launches on mobile and web for Max users
Claude Cowork is rolling out to mobile and web for Max users, with shared Chat navigation and cross-device task review. Scheduled tasks can run while a computer is off, and doubled Cowork limits last through August 5.
Anthropic extends paid Fable 5 access through July 12 with 50% weekly cap
Anthropic says paid Claude plans keep Fable 5 access until July 12, with Fable use capped at 50% of weekly limits. Users still reported exhausted quotas and multiple Max subscriptions as workarounds.
Meta launches Muse Image with reasoning, search, code, and self-refinement
Meta launched Muse Image in its apps and previewed Muse Video from the same media-generation family. Meta says Muse Image can reason, search, write code, self-refine, and use test-time compute before generating images.
OpenRouter benchmarks 1,730 visual-reasoning questions on low-detail image costs
OpenRouter tested 1,730 visual-reasoning questions across five models and found low-detail images often reduced accuracy while increasing reasoning-token spend. Caps on reasoning effort had the biggest billing impact.
Goodfire releases Block-Sparse Featurizers for DINOv3, SDXL, and InceptionV1
Goodfire introduced Block-Sparse Featurizers, which model activation concepts as multidimensional blocks instead of single SAE directions. The examples cover DINOv3, SDXL, and InceptionV1 activations.
ClaudeDevs reports Sonnet 5 + Fable 5 advisor hits ~92% SWE-bench Pro score at ~63% price
ClaudeDevs reports Sonnet 5 with a Fable 5 advisor reached ~92% of Fable 5's SWE-bench Pro score at ~63% of the price. Other builders route implementation to Sonnet, Codex, GPT-5.5, or GLM workers.
Top storiesthis week
AutomationBench-AA launches SaaS-agent leaderboard with 657 tasks
Artificial Analysis launched an independent Zapier AutomationBench leaderboard with 657 tasks across 40 simulated SaaS apps. Claude Fable 5 and Opus 4.8 led, but models still violated business guardrails.
Tencent releases Hy3, a 295B MoE model under Apache license
Tencent released Hy3 with 21B active parameters, a 256K context window, BF16/FP8 weights, and day-one vLLM/SGLang support. Kilo Code, Nous Portal, and OpenRouter also made it free for limited windows.
Anthropic claims Claude uses a small internal J-space activation workspace
Anthropic published a paper describing a small activation subspace where Claude represents concepts before text output, with demos for reading, editing, and ablating it. Researchers debated whether the method is novel or overstated.
Fable 5 users report heavy Claude Code credit burn before July 7 cutoff
A Fable maintainer set the Claude subscription cutoff for 11:59:59pm PT on July 7 as users posted high Claude Code credit usage. One estimate claimed about $2,267 of usage across two $200 accounts in six days.
LongCat-2.0 opens MIT weights for 1.6T MoE with 1M context
Meituan released LongCat-2.0 weights and inference code under MIT, with Hugging Face, GitHub, ModelScope, GPU, and NPU paths. Analysts noted the ~48B-active MoE keeps attention shape while reducing zero-communication experts from 256 to 128.