HealthBench Hard

OpenAI logoGPT-6.1 Terra on HealthBench Hard

rank 13 of 21 · updated November 2, 2026via Addendum to GPT-6 Astra System Card: GPT-6.1 Terra

On the 21-model HealthBench Hard board, GPT-6.1 Terra holds rank 13 with a score of 0.314. HealthBench Hard tests models on the 1,000 health conversations the frontier found hardest, with each response judged against its conversation's physician-written rubric and scored between 0 and 1.

Score and API facts

rank13 of 21
score0.314
sourceAddendum to GPT-6 Astra System Card: GPT-6.1 Terra (Section 5.1, Table 5, HealthBench Hard length-adjusted, GPT-6.1 Terra column), vendor-reported
configurationOpenAI; length-adjusted; 31.4 (26.0, 1306) (adjusted, raw, mean response characters); addendum Table 5 refers to the GPT-6 Astra card for method
labOpenAI
context window1.05M tokens
API price per 1M tokens$1.00 in / $5.00 out
licenseproprietary
released2026-10-26

Where it sits

Muse Spark tops the board at 0.428, which puts GPT-6.1 Terra 0.114 off the lead. One place up is GPT-6 Luna at 0.314. One place down is GPT-6 Sol at 0.301. Rows on this board are compiled from published documents, so a gap between two models is exact only when both numbers came from the same document under the same settings; the sources page shows which document each row came from.

What does GPT-6.1 Terra score on HealthBench Hard?

As of November 2, 2026, GPT-6.1 Terra scores 0.314 on HealthBench Hard, 13 of 21 models on the board. The number is read from the Addendum to GPT-6 Astra System Card: GPT-6.1 Terra (Section 5.1, Table 5, HealthBench Hard length-adjusted, GPT-6.1 Terra column), vendor-reported.

What does GPT-6.1 Terra cost per million tokens?

OpenAI lists GPT-6.1 Terra at $1.00 per million input tokens and $5.00 per million output tokens.

Head to head

The pairings that earned a full page are linked below; the rest of the differences live in the score-difference matrix.

How numbers are read from their documents is on the methodology page, the document for this row is on the sources page, and how the subset was selected is on the benchmark page. The full ranking is on the leaderboard.