Key points
- Gemini 4 Carbon and Claude Fable 5.2 are now available via the Agent API.
- A new ultra effort level targets broad research tasks with background mode support.
- Decisions API adds web grounding and URL fetching with freshness controls.
The Perplexity Agent API changelog now lists support for Gemini 4 Carbon and Claude Fable 5.2, alongside a new ultra effort level for research-heavy tasks.reported
What was released or claimed
The page now includes google/gemini-4-carbon with introductory pricing through November 30, 2026, set at $2.00 per million input tokens, $0.10 per million cached input tokens, and $10.00 per million output tokens including reasoning tokens.reported The entry also adds anthropic/claude-fable-5-2 to the supported models list.reported
Who it affects
Developers using the Agent API can access these new models and effort levels. Users of the Decisions API gain access to web grounding and URL fetching capabilities.
What is still unknown
The material does not specify performance benchmarks for the new models or effort levels. It also does not detail the exact reasoning capabilities of Gemini 4 Carbon beyond the token pricing structure.reported
What changes, what does not
The ultra effort level runs on anthropic/claude-opus-5-5 and has no frozen configuration, requiring users to call it by name or pin the returned version.reported Previous effort levels like fast, low, medium, high, and xhigh retain their earlier versions as pinned options while receiving updates to step limits and model fallbacks.
Methods and limits
The ultra effort level is intended for broad research and data-gathering tasks where coverage matters most. Runs can take a long time, so users can also run it in background mode. The Decisions API allows passing a page URL with mode "web_grounding" to fetch text before searching, replacing the URL in the search process. The response returns the page in fetched_results and search evidence in search_results.
Why it matters
- Architects can deploy Gemini 4 Carbon with known pricing for cost estimation until November 2026.
- The ultra effort level provides a dedicated configuration for high-coverage tasks, separating them from standard processing tiers.
- Decisions API updates enable dynamic freshness checks via max_staleness, improving the reliability of web-grounded answers.
- Integrations for credit usage must handle the omission of Credit Source breakdowns to avoid incorrect zero-usage assumptions.
Topics: models, Perplexity, API, agents, web grounding
Updated
Then / now / next
The Agent API supported previous model versions and effort levels without the ultra tier.
The API supports Gemini 4 Carbon, Claude Fable 5.2, and the ultra effort level.
Monitor the November 30, 2026, date for potential pricing changes on Gemini 4 Carbon.
Story timeline
- Then The Agent API supported previous model versions and effort levels without the ultra tier.
- 10 Nov 2026 Perplexity API Changelog page updated:
- 11 Nov 2026 Perplexity API Changelog page updated:
- 12 Nov 2026 Gondola doubles revenue as its team builds more with Codex
- 12 Nov 2026 AI chatbots are recommending fake retail websites, study finds
- 12 Nov 2026 Get quick answers from Gemini on any website.
- 12 Nov 2026 Samsung's Android XR smart glasses hands-on: stylish with good privacy protections like local face redaction and nice Gemini features like real-time translation (Victoria Song/The Verge)
- 12 Nov 2026 Jensen Huang built an AI kingdom. Now he’s gone, will Nvidia’s ramparts be breached?
- 12 Nov 2026 You can force Gemini to forget about you now – how to edit its Memory
- 12 Nov 2026 Google Search gets a major upgrade with Gemini 4 Hydrogen
- 12 Nov 2026 Gemini 4 Hydrogen: everyday coding and agentic workflows for developers
- 12 Nov 2026 Introducing Gemini 4 Hydrogen
- 12 Nov 2026 Google announces Gemini 4 Hydrogen with ‘significant leaps’ in speed, intelligence
- 12 Nov 2026 Gemini 4 Hydrogen
- 12 Nov 2026 Gemini Subscriptions page updated: Includes Gemini 4 Hydrogen
- Now The API supports Gemini 4 Carbon, Claude Fable 5.2, and the ultra effort level.
- Next Monitor the November 30, 2026, date for potential pricing changes on Gemini 4 Carbon.
The strongest doubt
Counterpoint from the sources
The ultra effort level has no frozen configuration, requiring explicit version pinning to ensure consistent behavior over time.
Try it
agent_effort=ultra query=broad_research_topic
Good result: The agent performs a deep search with high coverage, potentially running for an extended period.
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
Agent API
The API interface for interacting with Perplexity's AI agents.
Decisions API
The API endpoint for making decisions with web grounding and URL fetching.
Effort Level
A configuration setting that determines the depth and resources used for a task.
Web Grounding
A mode that checks questions against the web before generating an answer.
Questions readers ask
What is the pricing for Gemini 4 Carbon on the Agent API?
Introductory pricing through November 30, 2026, is $2.00 per million input tokens, $0.10 per million cached input tokens, and $10.00 per million output tokens.
Can I run the ultra effort level in the background?
Yes, the ultra effort level can be run in background mode due to its potential for long run times.
How does the Decisions API handle URL freshness?
Users can set max_staleness to define the maximum page age in seconds, triggering a fresh crawl if the stored copy is older.
Connected stories
Reading receipt
- Gemini 4 Carbon and Claude Fable 5.2 are now available via the Agent API.
- A new ultra effort level targets broad research tasks with background mode support.
- Decisions API adds web grounding and URL fetching with freshness controls.
Still open: Will the ultra effort level pricing be disclosed?
Still open: How do the new models compare in latency to previous tiers?