Perplexity CEO Aravind Srinivas has praised Gemini 4 Carbon as the strongest model Perplexity has tested so far for long-form research and answer quality, pointing to early evaluations and a new state-of-the-art result on the WANDR benchmark.
Perplexity · Aravind Srinivas
Perplexity's CEO praises Gemini 4 Carbon for research and answer quality
Early internal evaluations place Carbon ahead of other models Perplexity has tested, especially on long-form research. The headline claim is a new state-of-the-art result on WANDR, but the scores behind it have not been published.
What is claimed vs. what is available
CLAIMED
New state of the art on WANDR
Stated by the CEO as a headline result
NOT AVAILABLE
Benchmark scores
Size of the improvement cannot be measured
NOT AVAILABLE
Metrics and comparison models
How "better" was defined and against what
From early praise to a verdict: the steps still ahead
STEP 1 · DONE
Early evaluation by Perplexity
→
STEP 2 · OPEN
Wider testing beyond early conditions
→
STEP 3 · OPEN
Independent benchmark results
→
STEP 4 · OPEN
Broader user feedback
Why the endorsement matters
Comes from a company that tests models for live answer generation, not only academic benchmarks
Targets long-form research, a demanding use case that needs accuracy and coherence
Would support Carbon's competitiveness in research-focused AI products if it holds
Why caution is warranted
All results come from a single company's early evaluation
Improvement size and method cannot be independently confirmed
Unclear whether results reflect sustained production workloads
Bottom line
A strong endorsement with no published numbers yet.
The strength of the claim can only be judged once independent WANDR scores, the comparison models, and broader user feedback are available.
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article...