Review · AISource: Exa
Exa 3.0 adds lower-cost Lite search and generally available Dynamic Highlights
Exa 3.0 adds Lite, a search tier priced below Auto, to Exa’s web search and content retrieval API for AI applications. [1][2] Exa announced the update on October 19, 2026, and said it was immediately available through its API. [1] The company says Dynamic Highlights is now generally available. [1] It selects text across results to reduce duplication and allocate more space to useful sources, according to Exa. [1] Exa also says it added a larger embedding model, expanded coverage and built Lite to use Auto’s index with less retrieval compute. [1]
Web search API
Exa 3.0
VerdictWorth trying
Exa positions Exa 3.0 for AI applications that need compact web evidence and a lower base search price. [1][2]
- Best for
- Exa’s intended users include developers building web research agents and LLM applications with tight context and search budgets. [1][4]
- Pricing
- Lite: $4, Auto: $7, Deep: $12 per 1,000 requests, including up to 10 results per search; extras cost more. Free Starter includes $10 monthly credits plus a $10 onboarding bonus. [2]
Pros
- Lower base search price for Lite than Auto [2]
- Monthly free credits without a payment method [2]
- Cross-document text selection intended to reduce duplication, according to Exa [1]
- Upgraded highlights at no additional charge, according to Exa [1]
Cons
- Risk of omitted qualifications in extracts, according to Frontier AI Analysis [3]
- No guarantee of freshness from relevance alone, as Frontier AI Analysis cautions [3]
- Additional charges for extra results and content retrieval [2]
Why it matters
Selecting passages before they reach an LLM can help control context size. [1] Lite offers a lower base search price than Auto. [1][2] Brave Search API offers an independent index and filtering tools. [5] Tavily’s pay-per-call offering combines agent-oriented search with extraction, crawling and mapping. [6] Exa’s performance and token-saving figures are company claims. [1] Frontier AI Analysis warns that extracts can omit qualifications and recommends retaining underlying pages to check context. [3]
Attribution
Sources
- 1ExaIntroducing Exa 3.0
- 2ExaExa API Pricing
- 3Frontier AI AnalysisExa 3.0 shifts the search budget toward selected evidence
- 4ExaExa | Search API for AI Agents — Real-Time Web Data
- 5BraveBrave Search API
- 6TavilyTavily API — pay-per-call web search
Read the original source for full details and context.
Article · AI
vLLM 0.32.0 adds structured decisions: how to compare them with constrained JSON
vLLM 0.32.0 enables an endpoint that reads model scores for predefined answers. A proposed document-triage test compares that approach with constrained JSON across accuracy, calibration, latency, and fallback handling.
Article · AI
Build a Voxtral 3 booking prototype that checks corrections before committing
Mistral’s Voxtral 3 booking demonstration shows a corrected reference before a lookup. This guide proposes a prototype to test whether corrections also prevent obsolete bookings, with controls for delayed tools, confirmation and database writes.
Article · AI
Moving PyTorch to Trainium: How TorchNeuron’s three execution modes differ
TorchNeuron’s public beta brings eager execution, experimental adaptive eager and torch.compile to Trainium2 and Trainium3. This guide explains their compilation, precision and fallback behavior and outlines a controlled migration experiment.