• Sources: primary, discussion
  • Summary: Artificial Analysis reports Astra priced at 2.5 times GPT-5.6 Sol and measures a token reduction on coding tasks. The same write-up records regressions against Sol on other evaluations in its suite.
  • Why it matters: A team costing an agent workload gets a token reduction on coding tasks and a per-task price rise everywhere else, from an evaluator that also records where the model got worse.

send feedback on this story