Perplexity Launches Fast Search, Powered by Its In-House Photon Engine
On September 25, Perplexity released Fast Search, a new preset for its Search API, and announced that its in-house retrieval and ranking service, Photon, now fully powers retrieval and ranking across all of its products, replacing the previously used open-source search engine. The company says the switch dramatically boosts speed and cuts costs with virtually no loss in quality.
Confirmed
- Fast Search is powered by Photon, which is written in Rust and was built by a small team of engineers working alongside hundreds of AI agents
- Performance: 95% of search results return within 230ms, with a median latency of 160ms, which Perplexity claims is the lowest among all the search APIs compared
- Cost and quality: across six agent benchmarks, per-task cost drops 68% versus the default preset with comparable quality; on long-tail and broad-coverage queries, relevance scores are only 0.24 points lower
- Technical design: Photon stores its index in a compact format, reading and decoding only the data needed per query; disk fetches are batched; Fast Search spends less compute on ranking, making it well suited to most everyday agentic tasks
Why it matters
- Perplexity's move from relying on an open-source search engine to a fully in-house retrieval stack means greater control over its core retrieval pipeline and a better cost structure
- Fast Search targets the high-concurrency, low-cost needs of agentic workloads, putting it in direct competition with other search APIs
2026-09-25 ~ 2026-09-25 · 6 related posts
Primary sources
- Perplexity Launches Fast Search API: 95% of Results in Under 230ms on Rust-Based Photon — perplexity_ai ·
- Perplexity's in-house Photon engine now handles all retrieval, replacing open-source search — perplexity_ai ·
- Perplexity Fast Search cuts per-task cost 68% on six agent benchmarks at 160ms latency — perplexity_ai ·
- [source] Perplexity Launches Fast Search API: 95% of Results in Under 230ms on Rust-Based Photon — perplexity_ai · 2026-09-25
- [source] Perplexity Fast Search cuts per-task cost 68% on six agent benchmarks at 160ms latency — perplexity_ai · 2026-09-25
- Perplexity Fast Search trades 0.24 relevance points for much lower ranking compute — perplexity_ai · 2026-09-25
- [source] Perplexity's in-house Photon engine now handles all retrieval, replacing open-source search — perplexity_ai · 2026-09-25
2 near-duplicate retellings: perplexity_ai · soleio