Perplexity adds locate feature to Decisions API v1.3 without raising token prices
The new v1.3 models support span extraction with character offsets while keeping input costs at $0.02 per million tokens for the 27B model.
Key points
- pplx-decider-v1.3 and pplx-decider-flash-v1.3 are now available with unchanged pricing from v1.2.
- The new locate question type returns text spans with character offsets and confidence scores.
- Locate questions do not support web grounding and read only text, ignoring images.
Perplexity updated its API changelog to announce the arrival of pplx-decider-v1.3 and pplx-decider-flash-v1.3.reported These new decision models introduce a locate question type that extracts specific text spans from a provided state.The new models allow you to find specific text parts without paying more for input tokens.v1.3 introduces span extraction with offsets while maintaining v1.2 pricing and free output tokens.
What changed
The page now lists pplx-decider-v1.3 (27B) and pplx-decider-flash-v1.3 (4B) as the current decision models.reported Pricing remains at $0.02 per million input tokens for the larger model and $0.01 for the smaller one.reported Output tokens continue to be free. The aliases pplx-decider-latest and pplx-decider-flash-latest now point to these v1.3 versions.reported Users can still pin v1.2 explicitly if they need to maintain previous probability thresholds.reported
How it works
A locate question identifies parts of a string state that match a given description. The response includes the character offsets for each matching span along with a score. This feature is exclusive to the v1.3 models.reported Unlike other question types, locate does not work with web grounding. It processes only text, ignoring any images included in the input state. Multiple questions can be answered in a single pass, with each isolated from the others.
What the evidence means
The locate feature provides precise extraction capabilities without relying on external web searches. This limitation means users cannot combine span finding with real-time web data in a single request. Developers must revalidate any probability thresholds tuned on v1.2, as model updates can shift score distributions.reported The isolation of questions in one pass prevents cross-question interference during processing.
Cost and market position
Pricing stability is a key factor for product teams evaluating vendor choices. By keeping input costs at $0.02 and $0.01 per million tokens, Perplexity avoids forcing budget adjustments for users upgrading to v1.3.reported The free output tokens further reduce variable costs. This pricing strategy positions the Decisions API as a cost-effective option for high-volume text extraction tasks, competing on price parity with its previous version while adding new functionality.
Why it matters
- Product teams can extract structured data from text states without incurring additional output token costs.
- The lack of web grounding for locate questions simplifies compliance for teams avoiding external data sources.
- Stable pricing allows for predictable budgeting when migrating from v1.2 to v1.3.
- Revalidating thresholds is required to maintain accuracy in decision logic after the model update.
Topics: models, API, pricing, Perplexity, text extraction
Updated
Then / now / next
v1.2 models were the latest available decision models with their specific probability thresholds.
v1.3 models are live with locate questions and updated aliases pointing to them.
Monitor if future updates change the pricing or add web grounding support to locate.
Story timeline
- Then v1.2 models were the latest available decision models with their specific probability thresholds.
- 12 Nov 2026 Perplexity API Changelog page updated:
- 13 Nov 2026 Perplexity API Changelog page updated: Agent APIModels
- Now v1.3 models are live with locate questions and updated aliases pointing to them.
- Next Monitor if future updates change the pricing or add web grounding support to locate.
The strongest doubt
Counterpoint from the sources
The locate feature cannot process images, limiting its utility for multimodal workflows.
For readers
Developers
- Revalidate probability thresholds tuned on v1.2 due to model changes.
- Use character offsets to extract precise text spans from states.
Business
- Input costs remain stable at $0.02 and $0.01 per million tokens.
- Output tokens are free, reducing variable costs for high-volume usage.
Researchers
- Locate questions isolate multiple queries in a single pass.
- Web grounding is not supported for locate question types.
Ask the article
Local keyword search of this article; no generated answers.Did this article give you the evidence you needed?
Your choice stays on this device. No public tally.Terms in this story
locate question
A query type that returns text spans matching a description with offsets and scores.
web grounding
A feature that checks questions against the web before answering.
character offsets
The position of a text span within the original string.
state
The input text or data provided to the model for processing.
Questions readers ask
Can I use locate questions with web grounding?
No, locate questions do not work with web grounding and only process text inputs.
Does upgrading to v1.3 change my token costs?
No, input token prices remain $0.02 for the 27B model and $0.01 for the 4B model.
How do I stay on v1.2?
You must explicitly name pplx-decider-v1.2 or pplx-decider-flash-v1.2 in your requests.
Connected stories
Reading receipt
- pplx-decider-v1.3 and pplx-decider-flash-v1.3 are now available with unchanged pricing from v1.2.
- The new locate question type returns text spans with character offsets and confidence scores.
- Locate questions do not support web grounding and read only text, ignoring images.
Still open: Will web grounding support be added to locate questions in future updates?
Still open: How do the new models compare in accuracy to v1.2 for non-locate tasks?
