Search Intelligence Score
54.7
+32.3 lift from search
050100
- Lift from search
- +32.3
- 22.3 without search
- Cost per 1K tasks
- $23.0
- #1 in Search Efficiency
- Time per task
- 59.4s
- Search plus inference time to complete one task
Accuracy by benchmark
Accuracy on each eval, with search (solid) and without (light).
- DSQA52.0 vs 14.0
- HLE57.0 vs 42.0
- WISER55.0 vs 11.0
Cost per 1K tasks
How the cost splits between model inference, search calls, and page extraction.
- Inference
- $12.2
- 53%
- Search
- $6.89
- 30%
- Extract
- $3.95
- 17%
InferenceSearchExtract
Compared with
Other models from the same provider alongside models that score closest.
| Model | Score | Lift | Cost per 1K tasks | Time per task |
|---|---|---|---|---|
| GPT-5.6 Luna | 54.7 | +32.3 | $23.0 | 59.4s |
| GPT-5.6 SolGPT-5.6 Sol | 66.7 | +35.3 | $217 | 52.9s |
| GPT-6 (Astra)GPT-6 (Astra) | 70.3 | +35.3 | $301 | 34.7s |
| Gemini 3.7Gemini 3.7 Flash | 54.3 | +25.3 | $61.1 | 79.7s |
| GLM 5.3GLM 5.3 | 55.3 | +35.3 | $110 | 147s |
Search Intelligence Score averages accuracy on DSQA, HLE, and WISER with equal weight; cost and time are averaged the same way, with cost shown per 1,000 tasks. Latest update September 14, 2026. Full methodology.
# GPT-5.6 Luna · Search Capability Leaderboard
- Search Intelligence Score: 54.7 (#9 in Search Intelligence)
- Without search: 22.3 · Lift from search +32.3
- Cost per 1K tasks: $23.0 (#1 in Search Efficiency)
- Time per task: 59.4s
- Medals: Gold, Search Efficiency
- Lab: OpenAI · Model id: gpt-5.6-luna
## Accuracy by benchmark
| Suite | With search | Without search |
|---|---|---|
| DSQA | 52.0 | 14.0 |
| HLE | 57.0 | 42.0 |
| WISER | 55.0 | 11.0 |
## Compared with
| Model | Score | Lift | Cost per 1K tasks | Time per task |
|---|---|---|---|---|
| GPT-5.6 Luna | 54.7 | +32.3 | $23.0 | 59.4s |
| GPT-5.6 Sol | 66.7 | +35.3 | $217 | 52.9s |
| GPT-6 (Astra) | 70.3 | +35.3 | $301 | 34.7s |
| Gemini 3.7 Flash | 54.3 | +25.3 | $61.1 | 79.7s |
| GLM 5.3 | 55.3 | +35.3 | $110 | 147s |
Latest update September 14, 2026. Methodology: /leaderboard#methodology