Perplexity releases Decider v1.3 and Decider Flash v1.3 with text-span extraction and Apache-2.0 weights
Perplexity released Decider v1.3 and Decider Flash v1.3 on November 12, 2026, adding verbatim text-span extraction and downloadable Apache-2.0 weights. The company reports 91.6% mean accuracy for the larger model across 63 query-classification tasks.
Perplexity released Decider v1.3 and Decider Flash v1.3 on November 12, 2026, adding verbatim text-span extraction and downloadable Apache-2.0 weights, according to a Unite.AI report.
Key details
- Models: Decider v1.3 and Decider Flash v1.3.
- License: Apache-2.0, with downloadable weights.
- New feature: Locate, which returns exact matching text along with start and end character positions. Named-entity recognition is one of the tasks it covers.
- Accuracy: Perplexity reports 91.6% mean accuracy for the larger of the two models across 63 query-classification tasks.
- Comparison: The company says this is 6.54 percentage points above its production classifiers.
- Evaluation method: The references for the 63 tasks were independently relabeled by a frontier LLM.
The accuracy figures are Perplexity's own claims. The report does not name the frontier LLM used for relabeling.
What this means
Locate returns the matching text together with its character start and end positions. Extracted spans can therefore be tied back to exact locations in the original input. The Apache-2.0 weights are downloadable, so developers can obtain the models directly. The 91.6% mean accuracy is Perplexity's own result on references relabeled by a frontier LLM, and it is measured against Perplexity's production classifiers.
Related Articles
NVIDIA releases 2-billion-parameter Alpamayo 2 Edge weights on Hugging Face for autonomous driving
NVIDIA released the Alpamayo 2 Edge model weights on Hugging Face on November 10, 2026. The 2-billion-parameter model targets autonomous-driving developers working within onboard compute and latency limits, and ships with deployment recipes and evaluation tools.
Google launches Gemini 4 Hydrogen: 1M-token context, $1.50 input and $7.50 output per 1M tokens
Google announced Gemini 4 Hydrogen on November 12, 2026, with native 1 million-token input and up to 1 million output tokens. It costs $1.50 per million input tokens and $7.50 per million output tokens, and Google claims it scores 51 on the Artificial Analysis Intelligence Index at its default high thinking level.
Geodesic Intelligence releases NovaAtom-Lite, a biomolecule model it claims tops state-of-the-art systems
Geodesic Intelligence has released NovaAtom-Lite, an AI model that predicts the structures and interactions of biomolecules. Founder Quanquan Gu said the model topped a third-party benchmark in several drug discovery tasks, according to the South China Morning Post.
Baidu launches ERNIE Lightning reasoning model and ERNIE-Wizshot-ME video model on Qianfan
Baidu released ERNIE Lightning, a deep-thinking model built first for Baidu Search, and ERNIE-Wizshot-ME, an expressive video model, at Baidu World 2026 on November 12. Both are live on Baidu AI Cloud's Qianfan platform with APIs open to enterprise customers and partners. Baidu has not published benchmarks, parameter counts or context lengths for either.
Comments
Loading...