@ArtificialAnlys on Backscroll

5 tweets from @ArtificialAnlys that Backscroll ranked for substance.

Grok 4.5 is highly cost-effective - it costs $0.31 to run per Intelligence Index task, which is less than GLM-5.2 and Kimi K2.6, and a 5x lower cost than Claude Sonnet 5 (max) while performing better on the Intelligence Index. This places Grok 4.5 firmly on the performance/cost
0 likes · 0 retweets · 1 replies
While completing objectives, agents must also avoid breaking guardrails that represent business rules. Every model we evaluated at launch triggers guardrail violations, and violation-adjusted efficiency separates the leaders: Gemini 3.5 Flash completes 15.0 objectives per
1 likes · 0 retweets · 1 replies
AutomationBench-AA evaluates models on 657 workflow automation tasks developed by Zapier from real workflow patterns on its platform. Models orchestrate work across 40 simulated SaaS app environments via REST APIs, and every task is graded programmatically on the final state of
2 likes · 0 retweets · 1 replies
On First Partial Transcription, Universal-3.5 Pro Realtime in Max Accuracy mode achieves a 4.1% WER at 0.44s after end of speech, behind only ElevenLabs Scribe v2 Realtime (3.6%, 0.13s) on accuracy and ahead of Cartesia Ink-2 external endpoints (4.3%, 0.07s), though slower to
0 likes · 0 retweets · 1 replies
AssemblyAI has released Universal-3.5 Pro Realtime: a streaming Speech to Text model achieving 4.1% WER on AA-WER Streaming (~0.4s to first final), able to take in conversation context at the start of a call and after each agent turn, without reconnecting Universal-3.5 Pro
5 likes · 0 retweets · 1 replies

How it ranks: Backscroll reads my Twitter/X timeline and ranks every tweet for substance with an LLM taste rubric — not raw engagement. Curated by Surya Dantuluri.