Skip to content
Theorem Weekly TW markTWTheoremWeekly

The evidence behind the week in machine learning.

Model updates

Perplexity Agent API adds ultra effort level and new model support

The platform introduces a high-compute effort tier for research tasks while adding Gemini 4 Carbon and Claude Fable 5.2 to its available model roster.

Key points

  • Introductory pricing for Gemini 4 Carbon is set through November 2026, including specific rates for reasoning tokens.
Perplexity Agent API adds ultra effort level and new model support
Depth
10 outlets · newest source 89h ago

What was released or claimed

The changelog states that the Agent API now supports the ultra effort level, designated as [email protected]. This tier runs on anthropic/claude-opus-5-5 and lacks a frozen configuration. The documentation also adds google/gemini-4-carbon and anthropic/claude-fable-5-2 to the supported model list.

Evidence: reportedPerplexity API Changelog page updated: ​

pagewatch-perplexity

Evidence: reportedPerplexity API Changelog page updated: ​

pagewatch-perplexity

Evidence: reportedPerplexity API Changelog page updated: ​

pagewatch-perplexity

Who it affects

Machine learning engineers and developers using the Perplexity Agent API for automated research or data gathering.

What is still unknown

It also does not detail the specific capabilities of Claude Fable 5.2 beyond its availability.

Evidence: reportedJensen Huang built an AI kingdom. Now he’s gone, will Nvidia’s ramparts be breached?

aifleet-guardian

So what

Developers can now configure agents for longer-running research tasks with higher step limits. The inclusion of Gemini 4 Carbon with introductory pricing provides a cost-effective option for input-heavy workflows until late 2026.

Evidence: reportedPerplexity API Changelog page updated: ​

pagewatch-perplexity

What the paper shows

The documentation details that the ultra effort level is designed for broad research where coverage is prioritized over speed. The material notes that developers must call the model by name or pin the version since there is no frozen configuration.

Why it matters

  • The ultra effort level enables broad research tasks with background mode support for long-running executions.
  • Decisions API web grounding allows fetching page text before searching to ensure freshness and replace URLs with actual content.
  • Agent API effort levels have increased output token limits and step budgets, with high effort now supporting up to 40 steps.
  • Gemini 4 Carbon and Claude Fable 5.2 are now supported by the Agent API, expanding available model options.

Topics: models, vendors, tools, API, agents

Updated

Story timeline

  1. 10 Nov 2026 Perplexity API Changelog page updated: ​
  2. 11 Nov 2026 Perplexity API Changelog page updated: ​
  3. 12 Nov 2026 Gondola doubles revenue as its team builds more with Codex
  4. 12 Nov 2026 AI chatbots are recommending fake retail websites, study finds
  5. 12 Nov 2026 Get quick answers from Gemini on any website.
  6. 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)
  7. 12 Nov 2026 Jensen Huang built an AI kingdom. Now he’s gone, will Nvidia’s ramparts be breached?
  8. 12 Nov 2026 You can force Gemini to forget about you now – how to edit its Memory
  9. 12 Nov 2026 Google Search gets a major upgrade with Gemini 4 Hydrogen
  10. 12 Nov 2026 Gemini 4 Hydrogen: everyday coding and agentic workflows for developers
  11. 12 Nov 2026 Introducing Gemini 4 Hydrogen
  12. 12 Nov 2026 Google announces Gemini 4 Hydrogen with ‘significant leaps’ in speed, intelligence
  13. 12 Nov 2026 Gemini 4 Hydrogen
  14. 12 Nov 2026 Gemini Subscriptions page updated: Includes Gemini 4 Hydrogen

The strongest doubt

Counterpoint from the sources

The documentation warns that ultra runs can take a long time, which may impact latency-sensitive applications.

Ask the article

Local keyword search of this article; no generated answers.

Did this article give you the evidence you needed?

No vote saved on this device.
Your choice stays on this device. No public tally.

Questions readers ask

What is the pricing for Gemini 4 Carbon?

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.

Which model runs the ultra effort level?

The ultra effort level runs on anthropic/claude-opus-5-5.

Can I use web grounding in the Decisions API?

Yes, by setting mode to web_grounding, which runs a Fast Search for up to 100 questions before the model answers.

Connected stories

This storyDoes Qwen4-Max’s 2.4T MoE Architecture Deliver 50% Token Efficiency Gains?Does Qwen4-Max’s 2.4T MoE ArchitGemini page removes Project Genie note and Wallet integration reportGemini page removes Project GeniHow AlphaGenome 2 Predicts Cell-Specific DNA Variant EffectsHow AlphaGenome 2 Predicts Cell-Realtime-Bench evaluates AI systems on dynamic tasks where delayed actions affect successRealtime-Bench evaluates AI syst

Reading receipt

  • Introductory pricing for Gemini 4 Carbon is set through November 2026, including specific rates for reasoning tokens.

Still open: What are the specific performance characteristics of Claude Fable 5.2?

Still open: How does the ultra effort level compare in cost and speed to the xhigh tier?

Time spent: 0:00 · Depth: Full

Sources

  1. Perplexity page watch

Related stories

Background reading