@no_stp_on_snek: the cleanest thing the RL bought at 9B wasn't intelligence, it was efficiency. Ornith-9B lands the same answers in ~56%…
Summary
Ornith-9B demonstrates that RL training at 9B parameters primarily buys efficiency, achieving same answers with ~56% of the tokens and twice the speed of its base model, offering real cost savings for per-token payment.
View Cached Full Text
Cached at: 06/29/26, 10:34 AM
the cleanest thing the RL bought at 9B wasn’t intelligence, it was efficiency. Ornith-9B lands the same answers in ~56% of teh tokens its base needs (1,299 vs 2,312 per turn) and retries half as often. same destination, way less wandering, roughly 2x faster wall-clock. if you pay per token, that’s the headline.
Tom Turney (@no_stp_on_snek): verdict up front: it’s a “pass” in my book in certain categories, just a narrower one than the 35B. you’re buying real efficiency and sustained-coding coherence here, not a clean across-the-board upgrade over the base.
ran the same skeptic battery on Ornith-1.0’s little brother,
Similar Articles
@no_stp_on_snek: verdict up front: it's a "pass" in my book in certain categories, just a narrower one than the 35B. you're buying real …
The author evaluates Ornith-9B against its base Qwen3.5-9B, finding that RL post-training improves token efficiency and sustained coding coherence but sacrifices single-turn judgment and robustness to misleading inputs, making it a narrower upgrade at 9B compared to the 35B version.
@no_stp_on_snek: someone will wave the card at me: the 9B crushes its base on the coding benchmarks (SWE-bench 69 vs 53). true. but on m…
A commentator discusses the performance of a 9B model on coding benchmarks, noting that while it beats its base on SWE-bench (69 vs 53), the advantage narrows on behavioral and long-horizon tests, suggesting limited gains outside benchmark distributions.
@rohanpaul_ai: TokenPilot reduces LLM agent costs via ingestion-aware compaction and lifecycle-aware eviction. Achieves 61–87% cost re…
TokenPilot reduces LLM agent costs via ingestion-aware compaction and lifecycle-aware eviction, achieving 61–87% cost reduction on PinchBench and Claw-Eval with competitive scores.
Ornith-1.5-35B-A3B Q4 running 60tk/s on 4070Ti.
A user reports achieving 60 tokens per second with the Ornith-1.5-35B-A3B MOE model on an NVIDIA RTX 4070 Ti, demonstrating efficient local inference optimization.
A $500 RL fine-tune of a 9B open model beat frontier models on catalog review
A $500 RL fine-tune of a 9B open-source model achieves 87% catalog review quality at $0.50 per 1000 listings, significantly outperforming frontier models like GPT-4 and Claude which cost $19-$172 and achieve only 70-76% quality.