AI Updates on 2026-02-19

Google releases Gemini 3.1 Pro achieving 77.1% on ARC-AGI-2 (more than double Gemini 3 Pro's score) and leading Artificial Analysis Intelligence Index at less than half the cost of frontier peers @GoogleDeepMind
Gemini 3.1 Pro ranks tied #1 in Text Arena (1500 score), top 3 in Arena Expert (1538), and #6 in Code Arena on par with Claude Opus 4.5 @arena
Alibaba's Qwen3.5-397B-A17B becomes top 3 open model in Text Arena, ranking #20 overall and achieving 8.6x-19.0x faster decoding than Qwen3-Max @arena
Google launches Lyria 3 music generation model in Gemini App, creating music from ideas, images, or videos in seconds @JeffDean
Arcee.ai releases Trinity Large, first frontier-scale model in Trinity MoE family, now available in Text Arena @arena

OpenAI reportedly finalizing $100B funding deal at over $850B valuation @TechCrunch
World Labs raises $1 billion in new funding from AMD, Autodesk, Emerson Collective, Fidelity, NVIDIA, and Sea to unlock spatial intelligence @a16z
a16z leads Temporal's Series D as durable execution becomes critical for long-running AI agents at OpenAI, Replit, Lovable, and Abridge @a16z
Microsoft adds Grok 4.1 Fast to Copilot Studio's multi-model lineup for custom agent building @satyanadella
Linear migrates Oscar Health's 600 people from one of the world's most complex Jira instances in just one month by eliminating custom fields @GergelyOrosz

OpenAI commits $7.5M to AI Security Institute's Alignment Project to fund independent research on mitigations for safety and security risks from misaligned AI @OpenAINewsroom
Research finds AI models resist instructions to p-hack data but guardrails can be breached, raising alignment concerns for scientific misconduct @emollick
Nature Medicine study shows AI passed medical boards at 95% accuracy but when humans used it for triage, accuracy dropped below 35% versus Google control group @random_walker

Perplexity launches Comet iOS browser with pre-orders live, integrating AI assistance into every webpage with Safari-grade performance @AravSrinivas
Google Labs releases Pomelli tool that creates professional-grade marketing assets in seconds at no cost for small businesses @joshwoodward
PostHog introduces free log management with 50 GB monthly free tier at $0.25 per GB using OpenTelemetry with frontend and backend context @posthog
Cursor adds agent sandboxing on macOS, Linux, and Windows allowing agents to run securely and request approval only when stepping outside sandbox @cursor_ai

Gemini 3.1 Pro achieves 98% on ARC-AGI-1 at $0.52 per task and 77% on ARC-AGI-2 at $0.96 per task, pushing Pareto frontier of performance and efficiency @arcprize
François Chollet argues sufficiently advanced agentic coding is essentially machine learning, with optimization goals, search constraints, and blackbox outputs raising concerns about overfitting and concept drift @fchollet
NVIDIA releases Dynamo v0.9.0 with FlashIndexer achieving ~10B tokens/sec throughput and <10µs p99 latency on single node @NVIDIAAIDev
Microsoft Research releases comprehensive report on media integrity and authentication methods exploring practical paths toward trustworthy provenance across images, audio, and video @MSFTResearch