Bifrost's whole pitch is speed ("50x faster than LiteLLM"), and the benchmark is already open source, but the proof is split between a GitHub repo and scattered claims. The one page a developer choosing a gateway wants is a current, rerunnable head-to-head they can trust.
First action
Rerun the benchmarking repo on a documented instance and lock one set of numbers used everywhere.
How to measure it
Bifrost installs attributed to the benchmark page
- Now
- –
- Target
- +30% relative
- By
- 8 weeks
Steps
Rerun the benchmark
On documented hardware, current versions of all three.
Growth Marketer 1 day
Fix the number
One latency figure (not 20 and 11) across homepage, FAQ and docs.
Growth Marketer 2 hours
Build the page
P99, throughput, memory, success rate, with the reproduce command.
Growth Marketer 1 day
Show the method
Vegeta, mock provider, hardware, so critics can rerun it.
CTO 3 hours
Make it AI-readable
A markdown version in llms.txt so assistants cite current numbers.
Growth Marketer 2 hours
Seed it
Share in gateway and LLM-infra communities as a rerunnable benchmark.
Social Media Marketer ongoing
Why it works
Publish a single benchmark page that shows P99 latency, throughput, memory and success rate for Bifrost, LiteLLM and Portkey on identical hardware, dated, with the exact command to reproduce it and a link to the benchmarking repo. Resolve the 20 vs 11 microsecond discrepancy so the number is consistent everywhere.
The Growth Marketer owns comparison pages and is measured from organic traffic to activation. A reproducible benchmark is the most credible, most linkable, most AI-quotable asset for a performance product, and it converts the self-hoster straight into installing Bifrost.
Risks and how to handle them
Rivals dispute the benchmark
Publish the method and let anyone rerun it; that is the credibility.
Evidence
- Your websiteBifrost benchmarks are reproducible and public: the maximhq/bifrost-benchmarking repo runs Bifrost against LiteLLM and Portkey on latency percentiles, throughput and memory with Vegeta and a mock provider. Claims include P99 latency 54x faster, throughput 9.5x higher and success rate 11.22% higher than LiteLLM.Source
- Your websiteHomepage leads with Bifrost, "The Fastest Enterprise AI Gateway": 20 microseconds added latency, 5,000 requests a second, "50x faster than LiteLLM" on P99 latency, 68% lower memory, 1,000+ teams using Bifrost, 8+ providers and 1,000+ models. Enterprise pricing is not published; the call to action is "Request access" to a 14-day Bifrost Enterprise trial or "Book a demo".Source
- Competitor researchAI gateway rivals: LiteLLM (the open-source default proxy, weak on analytics and caching), Portkey (governance-first; reports of a Palo Alto Networks acquisition in 2026), OpenRouter (hosted, pass-through plus a markup), Cloudflare AI Gateway (managed, ~20 providers), and Kong AI Gateway (for teams already on Kong). Bifrost competes on raw speed and being self-hostable and open source.Source
- Your websiteThe homepage and FAQ disagree on Bifrost's added latency (20 microseconds vs 11 microseconds), and the homepage carries unnamed customer logos; the careers and homepage both use "Ship your AI agents 5x faster" and claim teams save "100s of hours".Source