April 1, 2026
Cheap brains, big claims, bigger debates
Trinity Large Thinking
A big, open, cheap AI drops—patriots cheer, testers say “meh”
TLDR: Arcee AI released a cheap, downloadable reasoning model with a free trial, thrilling open‑source and USA‑proud fans while drawing skepticism over a 16/25 benchmark score versus rivals. The fight is on between price-and-openness hype and “numbers don’t lie” testers, with real‑world value still up for debate.
Arcee AI just dropped Trinity Large Thinking, a downloadable, open reasoning model with a jumbo memory and bargain pricing ($0.25 per million input tokens, $0.90 per million output), plus a free five‑day promo and a launch video. You can even toggle a setting to see its step‑by‑step thinking. The community? Instantly split between deal fever and data drama.
The hype crew is loud: one commenter called it “crazy affordable,” even hinting it might challenge today’s hot rival, StepFun 3.5 Flash (HN link). Another cheered that this is “one of the first high‑performing fully open weight American models,” complete with the meme‑y “(insert American flag here).” And yes, the weights are actually up on Hugging Face, surprising a few skeptics. For official details, a helpful soul dropped the Arcee blog.
Then the bucket of cold water: a community benchmarker reported Trinity scoring 16/25 on their SQL test, noting even the preview only hit 17/25, while a rival model scored 23/25. Cue the clash: “cheap and open” vs “show me the numbers.” Fans say cost and openness will win; skeptics say performance is “mid.” Expect more head‑to‑head smackdowns in the cost arenas as everyone tries to figure out if this is a budget hero or just loud marketing.
Key Points
- •Arcee AI released Trinity Large Thinking, an open-source reasoning model with a 262,144-token context window.
- •Pricing is $0.25 per million input tokens and $0.90 per million output tokens.
- •The model is described as strong on PinchBench, agentic workloads, and reasoning tasks.
- •OpenRouter provides provider routing, performance comparisons, effective pricing tracking, usage metrics, and uptime stats for the model.
- •OpenRouter’s API supports a “reasoning” parameter and exposes reasoning_details for step-by-step reasoning; developers should preserve reasoning_details across turns.