SuperCosmos 1.1
SuperCosmos Heavy
The deliberate tier. Thinks longest and writes the longest answers, for the hardest problems.
Heavy spends the most effort per step. It's slower and costs more, and it's the tier to reach for when the other two can't crack a problem.
| Model id | supercosmos-heavy |
| Model | SuperCosmos: 550B parameters, 55B active per token |
| Context window | 1M tokens (1,048,576) |
| Agent API | "tier": "heavy" |
| Input | $2.00 per 1M tokens |
| Output | $4.40 per 1M tokens |
| Long context | Long-context input above 262K tokens: 1.5× |
Good fits
- Hard bugs: race conditions, memory issues, flaky tests
- Large builds and migrations
- Architecture and design reviews
- Tasks where a wrong answer is expensive
Heavy is rarely the right default. Use auto or Flash for most runs and send only the hard ones to Heavy.
Was this page helpful?