In a 460-task BBEH Mini evaluation, SpaceXAI’s Grok 4.6 led the field with a 0.67 score at $0.0034 per task. Anthropic’s Claude Opus 4.8 followed at 0.62, while OpenAI’s GPT-5.6 Sol Pro posted 0.59.
By RuntimeWire Staff · Published
Leaderboard
| Rank | Model | Mean score | Est. cost/task | | 1 | SpaceXAI: Grok 4.6 | 0.67 | $0.0034 | | 2 | Anthropic: Claude Opus 4.8 | 0.62 | $0.0485 | | 3 | OpenAI: GPT-5.6 Sol Pro | 0.59 | $0.0089 |
How we scored it
Every model answered the same 460-task battery from BBEH Mini (BBEH — Google DeepMind (Apache-2.0)), one task at a time, with no tools and no retries on content.
460 tasks carry ground truth (exact, numeric, multiple-choice, or the benchmark's official matcher) and were scored deterministically against the reference answer — correct is 1, incorrect is 0.
A model's mean score averages its graded tasks; a generation failure counts as 0. Cost per task is estimated from each model's published per-token pricing ("—" where pricing isn't public), so treat it as directional, not billing-exact.
Explore every prompt, answer, and per-task grade in the interactive leaderboard.
Reader comments #
Conversation for this story loads after sign-in.