News archive
October 13, 2026
Meta releases Muse Wonder for personal and professional AI agents
Meta's Muse Wonder is available through its Model API on October 12, with Wonder priced at $4 per million input tokens and $20 per million output tokens; builders can compare it with their current agent model.
$4/M Wonder input tokens$20/M Wonder output tokens2 models in the release
Helix World generates interactive video and sound together
Helix World runs at 24 frames per second with a reported 0.77 real-time factor; makers can try the browser preview and test whether a generated scene stays coherent after repeated actions.
24 fps reported H800 speed0.77 reported real-time factor41.8 HelixBench spatial-acoustic score
Anthropic says new controls cut model-distillation data by 90% to 99%
Anthropic reports a 90% to 99% reduction in harvested data after its October 12 defenses; API teams should expect more account checks and less access to reusable reasoning traces.
90-99% reported reduction in harvested dataAug 31 cutoff for new API accounts3 defense methods described
LiteLLM 1.105.0 adds a Decisions API and agent gateway helpers
LiteLLM v1.105.0 shipped on October 11 with a Decisions API across eight providers and litellm.agent(); teams can test one gateway before routing more traffic.
1.105.0 release version8 providers named for Decisions API2 decision routes listed
BetterWispr brings free local dictation to macOS
BetterWispr launched on Hacker News on October 11 as a free Apache 2.0 macOS 14+ app with Option-Space dictation, local speech models and no required account.
macOS 14+ minimum systemApache 2.0 licence3 local speech families named
A 500B-token game decompilation project reaches 83% byte-exact code
The project reports 99% of functions present and 83% byte exact after nearly 3 months, using 14 Luna and 2 Opus agents; the useful idea is the acceptance harness, not the token bill.
83% functions reported byte exact99% functions present in source16 agents named in final weeks
ngram 2.0 turns product documents and recordings into narrated videos
ngram 2.0 ranked #2 on Product Hunt on October 12 with 226 points; makers can turn a product recording into a narrated walkthrough and refine it in chat.
#2 Product Hunt daily rank226 Product Hunt points3 video modes named
guilds.run opens inspectable AI agent teams as open source
guilds.run launched on Product Hunt on October 12 with a Docker setup and AGPL code; builders can inspect every prompt, tool call and memory change in a local agent team.
3 setup commands in the READMEAGPL-3.0 software licence1 local operator UI
Blueprints AI says it can generate permit-ready plans in minutes
Blueprints AI says its public v2 release begins the week of October 13 and can generate several plan disciplines at once; its first version was priced at $3,200 to $8,500 per month.
v2 public release version$3,200-$8,500 reported first-version monthly price8 years approved drawing sets used for training
Honeyjar automates PR research but keeps the send button human
Honeyjar says its waitlist gained 10,000 signups in 24 hours; the October 12 beta drafts PR work but does not publish or send without a customer.
10,000 reported first-day signups24 hours time to reach that waitlist count4 PR tasks named in the report
Ambi raises $26M for an ambient assistant across iPhone, Mac and Watch
Ambi has raised $26M across rounds and is building iPhone, Mac and Apple Watch apps; the company says ambient listening is off by default and a hardware device is planned.
$26M reported funding>50 reported staff3 current device classes
Anthropic admits queued Pro and Max users to Claude Code Projects beta
Anthropic says queued Pro and Max users entered Claude Code Projects by October 12, while the waitlist stays open and future access depends on capacity; builders can test parallel task threads if admitted.
2 paid plans named1 beta waitlist0 public capacity or date figures
Perplexity's pplx-decider v1.1 lifts Decision Index to 61.56
Perplexity's pplx-decider v1.1 raised its Decision Index score from 56.4 to 61.56; the 26B model needs authenticated Hugging Face access and about 49 GiB for weights plus working memory.
61.56 reported Decision Index score26B model parameters49 GiB reported weights plus working memory
Dropstone CLI launches a self-learning coding agent for the terminal
Dropstone CLI launched on Product Hunt on October 12 with a free plan and paid plans reported from $5 to $150; it keeps corrections as lessons across coding sessions.
87 Product Hunt launch points$5-$150 reported paid plan range3 model choices named
Epic reportedly tells staff to move toward AI-first game production
Ars reports a $1.5B Epic-Disney collaboration and internal pressure toward AI-generated code and art; a survey cited in the report says 36% of game professionals use generative AI.
$1.5B reported Epic-Disney collaboration36% game professionals using generative AI30% reported studio use
Pensieve gives coding agents local search across files and past sessions
Pensieve reports 43% fewer tool calls and 18% lower lookup cost in a 144-run benchmark; builders can test local context search before adding a hosted retrieval service.
43% reported fewer tool calls18% reported lower lookup cost144 agent runs in the benchmark
Amazon says machine learning improves a million-pick robot test
Amazon's reported picker test reached 93.73% accuracy with machine learning versus 92.28% for a traditional method, or about 16,500 fewer failed picks in one million.
1M robots in Amazon's network93.73% reported ML picking accuracy16,500 fewer failed picks in one million
