Fable 5.2 is twice the token price of Opus 5.5. that's the wrong number to stare at.
Anthropic says Fable at High matches Opus at Max on FrontierCode 2.0 for about half the cost per task. Claude's own team also says Max is for finding the ceiling, not everyday use.
so the real decision is how much effort each job deserves.
a research agent running for hours might need xhigh. a code review probably needs medium. a simple workflow might still belong on Opus.
the next edge in AI ops is effort routing: every task gets the cheapest setting that can actually finish it.
Introducing Claude Fable 5.2 and Claude Mythos 5.2, designed to tackle the most daunting technical challenges.
They’re capable collaborators on ambitious projects in many fields, and show improvements in areas like coding, reasoning, and vision.
56
GPT Image 2.6 is already looking like a big jump. the left version is the first time i've seen an image model treat 10 styles as 10 different worlds, not 10 filters on the same face
left is the "sunburst" checkpoint, right is "shooting-star". both were tested on arena before 2.6 showed up in openai's docs
credits to @chetaslua for the side by side images.
1
170
Meta says Muse Spark Open is the most capable US open-weight model yet. 816B total, 42B active, 1M context.
Their own chart has it leading on agentic and multimodal, but Kimi K3 and GLM 5.3 still win on coding.
The US is finally back in the open-weight race.
58
new GLM model would be huge. GLM 5.3 Flash has been working wonders. If they release a better model than that, I have to switch up fast
234
SpaceXSI says Grok Voice 3.0 answers in under a second, with 40% lower latency than GPT-Live-1 on support-call tests.
They also say Starlink uses Grok Voice for 20K support calls a day, with about 70% resolved with no human.
Call centers should be paying attention.
34
A simple rule from this week's Gemini 4 Argon demos: an AI agent that can check its own output beats one that just answers. Before you try a new model, set up the feedback loop first.
51