right signal
Live Reviewed on 13 Sep 2026, 14 September

What’s actually
best in AI.
Right now.

One current pick for every job.
Independently evidenced. Changed when the evidence changes.

Explore the picks What’s changing
10Current picks
Holding their titles
8Contenders on radar
Could change the board

The picks

Our current answer for every category.

View all picks
Language
Overall LLM Claude Fable 5.1 Pick4 days Coding GPT-6 Astra Pick5 days Open weight GLM-5.3 Pick4 days On device Gemma 4 26B A4B Pick165 days
Speech
Speech → text ARK-ASR-3B Pick84 days Text → speech BreezeBlue Breeze TTS 2 Open Weights Pick5 days
Create
Image GPT Image 2 Pick146 days Video Gemini Omni Flash Pick76 days
Build
Embeddings Nemotron-3-Embed-8B Pick60 days Inference vLLM Pick1182 days
Detailed reasoning, benchmarks and history for every pick. See all picks

On the radar

New and emerging contenders that could change the board.

See all 8 contenders
We track new releases and early data so you see what could change next. How the radar works

What changed

The most recent title changes and notable events.

View full changelog
13 SEP Wan 3.0's lead over Gemini Omni Flash widens to five Elo on AA's with-audio text-to-video boardArtificial Analysis's with-audio text-to-video board, read today, puts Alibaba Wan 3.0 first on 1242 with Gemini Omni Flash second on 123… Challenge 13 SEP Breeze TTS 2 slips to #9 as StepAudio 2.5 TTS edges aheadThe pick has lost ground. Challenge 13 SEP Weekly review: all 9 titles re-verifiedAll 9 picks were re-verified against the live leaderboards, repositories and release pages. Hold

Why RightSignal?

We do things differently.

One answer

We make the call. No endless leaderboards — just the best choice right now.

Evidence over hype

A vendor announcement isn’t enough. We wait for independent evidence.

Every change recorded

The full history of every title — what changed, why, and what replaced it.

Continuously reviewed

Reviewed daily, verified weekly. Titles change when the evidence changes.

How we judge and test models