Claude Opus 5.5 went absolutely off the rails for me. This is my version of hacking hugging face, this cost $10 and is ALSO ABSURD.
I was simply fixing a bug in the mobile display of a chart using OpenCode and Opus 5.5. The incentive no one talks about is the incentive to use more tokens.
While people are debating if now is the time to short @AskVenice base:0xacfe6019ed1a7dc6f7b508c02d1b04ec88cc21bf, they are signing up massive amounts of new paying subscribers, over 2500 per day this week so far. Think about the scale of that business. I’m long, not financial advice. These data are sourced from their onchain data, and may not be 100% accurate because my agent created them.
I've been playing with Jev from @typesafeai using @AskVenice for other text models. It blows away SOTA text models on cost and performance. Thanks for the access @notkevinzhang! @jesseproudman Venice should add this!
Has anyone else seen this in Opus 4.8? I think this is an artifact of @AnthropicAI's new security features. Opus completely fabricated a hack, scared the crap out of me until I dug in.
@volokuleshov - I think Mercury 2 would be one of the BEST models out there if it were better at tool calling. Here's a benchmark I've run on it - essentially it should be calling set_travel_dates 100% of the time, really you get 90%. I hope this is an area of focus, I love what you're doing!!!
I used Venice’s agentic chat to annotate this image a bit, showing:
• Top line = absolute new tokens added that month (in G)
• Bottom line = percentage growth vs prior month
Why are passive index investors being forced to buy SpaceX right out of the gate? This seems like a fucking scam to me. Ma and Pa gotta provide liquidity to Sequoia?
I’ve taught my @Agent0ai agent to do benchmarks of inference providers. Here is GLM 5.1 on @AskVenice vs ZAI.
