Note
Access to this page requires authorization. You can try signing in or changing directories.
Access to this page requires authorization. You can try changing directories.
Microsoft Web IQ is designed to deliver high-quality grounding with low latency and fewer tokens. The following benchmark results compare Web IQ with other grounding services.
Token efficiency
Web IQ operates on a new Pareto curve of efficiency. It prioritizes the most relevant passages and minimizes unnecessary context. This approach reduces token usage per query, reasoning time, and the total cost of delivering high-quality responses.
In the measured 3,000-query set, Web IQ delivered the highest-quality answers with the fewest tokens relative to the compared services.
Source: 3,000-query global production sample. Web result configurations included 10, 15, and 20 results, with 3,000, 5,000, 10,000, and 20,000 characters per result or the equivalent.
Grounding satisfaction
Web IQ delivers complete, structured context from web pages, news, images, and video. This context improves the accuracy and consistency of AI-generated responses.
| Service | Grounding satisfaction |
|---|---|
| Web IQ | 79.05 |
| Competitor A | 75.70 |
| Competitor B | 75.61 |
| Competitor C | 73.96 |
| Competitor D | 72.89 |
| Competitor E | 67.53 |
| Competitor F | 64.71 |
Source: 3,000 global blind queries sampled from production. The configuration used 10 results and 10,000 characters per result or the equivalent.
P95 latency
Fast grounding helps AI systems remain responsive across multi-step agent chains and dynamic queries. Reducing latency at each step compounds performance improvements across complex interactions.
| Service | P95 latency in milliseconds |
|---|---|
| Web IQ | 164 |
| Competitor C | 406 |
| Competitor E | 463 |
| Competitor A | 726 |
| Competitor B | 1,007 |
| Competitor D | 1,930 |
| Competitor G | 2,090 |
Web IQ's measured P95 latency was approximately 2.5 times faster than the next service in the comparison.
Source: Virtual machines hosted in West US 2, North Central US, East US 2, North Europe, and South Korea. P95 values are averaged across datacenters. Unique queries avoided cache hits. The configuration used 10 results and 10,000 characters per result.