HealthBench Hard

Compare models

updated November 2, 2026

Matchups that decide deployments get a full page each, with the score gap, list prices, and a costed usage workload. The matrix below covers every remaining pairing.

Head-to-head pages

Score-difference matrix

Each cell is the row model's score minus the column model's, so positive means the row wins. Column numbers follow rank.

model123456789101112131415161718192021
1. Muse Spark·+0.008+0.062+0.066+0.081+0.085+0.097+0.101+0.108+0.113+0.114+0.114+0.114+0.127+0.128+0.137+0.141+0.169+0.174+0.199+0.320
2. GPT-5.2-High-0.008·+0.054+0.058+0.073+0.077+0.089+0.093+0.100+0.105+0.106+0.106+0.106+0.119+0.120+0.129+0.133+0.161+0.166+0.191+0.312
3. GPT-6 Astra-0.062-0.054·+0.004+0.019+0.023+0.035+0.039+0.046+0.051+0.052+0.052+0.052+0.065+0.066+0.075+0.079+0.107+0.112+0.137+0.258
4. GPT-6.1 Sol-0.066-0.058-0.004·+0.015+0.019+0.031+0.035+0.042+0.047+0.048+0.048+0.048+0.061+0.062+0.071+0.075+0.103+0.108+0.133+0.254
5. GPT-5-0.081-0.073-0.019-0.015·+0.004+0.016+0.020+0.027+0.032+0.033+0.033+0.033+0.046+0.047+0.056+0.060+0.088+0.093+0.118+0.239
6. GPT-5.2-0.085-0.077-0.023-0.019-0.004·+0.012+0.016+0.023+0.028+0.029+0.029+0.029+0.042+0.043+0.052+0.056+0.084+0.089+0.114+0.235
7. GPT-5.6 Sol-0.097-0.089-0.035-0.031-0.016-0.012·+0.004+0.011+0.016+0.017+0.017+0.017+0.030+0.031+0.040+0.044+0.072+0.077+0.102+0.223
8. GPT-5.6 Terra-0.101-0.093-0.039-0.035-0.020-0.016-0.004·+0.007+0.012+0.013+0.013+0.013+0.026+0.027+0.036+0.040+0.068+0.073+0.098+0.219
9. GPT-5.6 Luna-0.108-0.100-0.046-0.042-0.027-0.023-0.011-0.007·+0.005+0.006+0.006+0.006+0.019+0.020+0.029+0.033+0.061+0.066+0.091+0.212
10. GPT-5.5-0.113-0.105-0.051-0.047-0.032-0.028-0.016-0.012-0.005·+0.001+0.001+0.001+0.014+0.015+0.024+0.028+0.056+0.061+0.086+0.207
11. GPT-5.6 Sol (August)-0.114-0.106-0.052-0.048-0.033-0.029-0.017-0.013-0.006-0.001·0.0000.000+0.013+0.014+0.023+0.027+0.055+0.060+0.085+0.206
12. GPT-6 Luna-0.114-0.106-0.052-0.048-0.033-0.029-0.017-0.013-0.006-0.0010.000·0.000+0.013+0.014+0.023+0.027+0.055+0.060+0.085+0.206
13. GPT-6.1 Terra-0.114-0.106-0.052-0.048-0.033-0.029-0.017-0.013-0.006-0.0010.0000.000·+0.013+0.014+0.023+0.027+0.055+0.060+0.085+0.206
14. GPT-6 Sol-0.127-0.119-0.065-0.061-0.046-0.042-0.030-0.026-0.019-0.014-0.013-0.013-0.013·+0.001+0.010+0.014+0.042+0.047+0.072+0.193
15. GPT OSS 120B-0.128-0.120-0.066-0.062-0.047-0.043-0.031-0.027-0.020-0.015-0.014-0.014-0.014-0.001·+0.009+0.013+0.041+0.046+0.071+0.192
16. GPT-5.4-0.137-0.129-0.075-0.071-0.056-0.052-0.040-0.036-0.029-0.024-0.023-0.023-0.023-0.010-0.009·+0.004+0.032+0.037+0.062+0.183
17. GPT-5.6 Luna (August)-0.141-0.133-0.079-0.075-0.060-0.056-0.044-0.040-0.033-0.028-0.027-0.027-0.027-0.014-0.013-0.004·+0.028+0.033+0.058+0.179
18. GPT-5.3 Chat-0.169-0.161-0.107-0.103-0.088-0.084-0.072-0.068-0.061-0.056-0.055-0.055-0.055-0.042-0.041-0.032-0.028·+0.005+0.030+0.151
19. GPT-5.1-0.174-0.166-0.112-0.108-0.093-0.089-0.077-0.073-0.066-0.061-0.060-0.060-0.060-0.047-0.046-0.037-0.033-0.005·+0.025+0.146
20. GPT-5.5 Instant-0.199-0.191-0.137-0.133-0.118-0.114-0.102-0.098-0.091-0.086-0.085-0.085-0.085-0.072-0.071-0.062-0.058-0.030-0.025·+0.121
21. GPT OSS 20B-0.320-0.312-0.258-0.254-0.239-0.235-0.223-0.219-0.212-0.207-0.206-0.206-0.206-0.193-0.192-0.183-0.179-0.151-0.146-0.121·

On a 1,000-conversation set, anything within 0.01 sits inside the variation between two measurements of the same model; read those cells as ties. Rows read from different documents carry the extra caveats on the methodology page; the document behind each row is on the sources page.

Pricing and context windows for every model are on the models page.