REVIEWS / AI SEARCH / OWNER INSIGHTS

🦉 WE READ 91 OWNER COMMENTS

Phind: what owners actually say

Owners find Phind faster than GPT-4 and great for obscure searches, but it hallucinates confidently and its 'beats GPT-4' marketing claim doesn't hold up in real testing.

HACKERNEWS · 75 LEMMY · 11 PRODUCTHUNT · 5

What owners complain about

  • Confident hallucinations COMMON

    Multiple users report Phind asserting things that don't exist with full confidence — e.g., claiming Cisco LISP supports map server database replication when it does not, or 'bullshitting' when it couldn't implement blocked clause decomposition in Haskell instead of admitting it couldn't.

  • Doesn't actually beat GPT-4 at coding COMMON

    Users who tested Phind's claim of beating GPT-4 found it underperforms on real tasks — e.g., a PostgreSQL query task where GPT-4 outperformed. One user called the benchmark claim 'deceptive' since it didn't actually include GPT-4 in comparisons.

  • Inconsistent code generation SOME

    Users report enabling certain options broke code generation entirely — one user's LeetCode-style prompt produced no code. Another found it failed at a typical CI/CD YAML generation task for a multi-project repo.

  • Pricing structure feels off FEW

    One user questioned an 8x price-per-search difference between Plus and Pro tiers, interpreting it as either Plus users being overcharged or Pro being unsustainably subsidized.

  • Dark mode readability FEW

    Dark mode has font weight and brightness issues making long non-code paragraphs hard to read; light mode is fine.

What owners love

  • Superior to traditional search

    After hours failing to find an obscure 8-pin GPU connector for wave-soldering on major electronics distributors, Phind surfaced the answer. Users describe it as a better search engine than Google, Kagi, or Bing for technical queries.

  • Fast response times

    Users note Phind runs noticeably faster than GPT-4, with one citing up to 100 tokens per second single-stream versus GPT-4's ~20 tokens per second.

  • Eager specificity

    For high-level system design questions (e.g., distributed job queue architecture), Phind was praised for proactively recommending specific technologies and approaches rather than giving generic advice.

  • Accessible for non-expert programmers

    A self-described 'lousy programmer' doing IT automation in Python/PowerShell found Phind genuinely useful for scripting tasks, suggesting it serves non-primary developers well.

Surprising patterns

  • Search mode can actually perform worse than Chat mode on the same question — one user found this with an IoT device authentication query, where Chat gave a better answer than Search.
  • Multiple highly upvoted comments are skeptical of Phind's benchmark methodology, suggesting the community sees coding benchmarks as potentially contaminated or gamed rather than reflecting real-world performance.
  • Users note Phind takes more internal generations to reach correct answers on hard questions, meaning its speed advantage may come with an accuracy trade-off on complex problems.

WHO SHOULD SKIP IT

Buyers who need reliably accurate answers for niche technical questions or existence-checks (where the correct answer is 'this doesn't exist') should skip Phind, as it tends to confidently fabricate rather than admit uncertainty.

5.8/10 GYIBB verdict
Full review →

Synthesised from 91 real owner comments across 3 platforms. Every point is grounded in the comments — no marketing, no AI guessing. How we do it →