31 articles
Browse guides
Ox Alpha is an anonymous frontier-class AI model on OpenRouter and OpenCode: 1M-token context, video input, free for roughly a week. The specs, early tests, identity theories, and how to try it safely.
OpenAI’s unreleased Astra model reportedly solved ten open math problems, including a non-sofic group, and published verifiable Lean proofs. The claims, caveats, and why it matters.
A June 2026 US executive order created a pre-release review process for ‘covered frontier models,’ with a classified threshold and a ~30-day access window. What it does, what stays secret, and why it matters.
DeepSeek V4-Flash-0731 is an official, open-weight, MIT-licensed MoE model with a 1M-token context, tuned for agents and coding, at about $0.14 per million input tokens. Specs, benchmarks, and why open weights plus low prices pressure the closed labs.
OpenAI cut GPT-5.6 Luna 80% on input three weeks after launch, part of a broad AI price war driven by cheap open-weight rivals. What changed across Sol, Terra, and Luna, and how to cut your AI costs.
In late July 2026, Claude’s public share links were missing a working noindex instruction, so Google and Bing began indexing shared conversations — some with highly sensitive data. Anthropic added the missing tags and Google started removing results by July 26. Nothing was hacked; this was a consent-and-configuration failure, and the fix on your side is to review and revoke your shared links.
Claude Mythos 5 is Anthropic’s restricted-access Mythos-class model — the same underlying model family as Fable 5 but with fewer safeguards and stricter access controls. It is limited to approved Project Glasswing and trusted-access partners. This guide covers what Mythos 5 is, how access works, benchmark data, safety design, and when Mythos-class capability matters versus Fable 5 or Opus 5.
Every Claude model and pricing tier in one guide. API pricing for Fable 5 ($10/$50), Opus 5 ($5/$25), Sonnet 5 ($3/$15 intro $2/$10), and Haiku 4.5 ($1/$5) per MTok. Subscription plans from Free to Enterprise. Prompt caching, batch API, fast mode, and cross-provider comparisons — all verified July 25, 2026.
Polymarket reports Grok 4.5 has overtaken GPT-5.6 Sol and Fable 5 in OpenRouter token volume, cracking the top 10 closed models. The milestone measures usage, not quality — but it lines up with a genuinely aggressive value proposition: $2/$6 per 1M tokens, roughly 2x token efficiency, 80 TPS, a SWE Marathon lead, and Cursor-co-trained coding performance. Here is what the signal means, what xAI’s benchmarks show, and what they do not.
Black Forest Labs launched FLUX 3 on July 23, 2026, in Early Access. Unlike FLUX 1 and FLUX 2, which generate images, FLUX 3 is a single multimodal flow model trained jointly on images, video, and audio — and even robot actions. It generates video with native audio up to 20 seconds, and an early version already powers FLUX-mimic, a video-action robotics model deployed at Audi. Here is what FLUX 3 does, what is still preliminary, and what the rollout plan covers.