Claude Opus 5.5 now available

We had early access to Claude Opus 5.5 ahead of its public release, and as of today it's live for every Saga user.

The benchmark that actually matters for legal work isn't the coding one everyone's been sharing this week. Anthropic tests its models on GDPval-AA, a benchmark built specifically around real-world professional output across fields like law, finance, and research, and Opus 5.5 scores 1846 Elo there against Opus 5's 1708, one of the larger single-version jumps on that measure to date. It's also 30% faster than its predecessor.

In our own early access period, that showed up as visibly better first-draft quality on exactly the kind of document review, gap analysis, and long-context contract work that already makes up most of what runs through Saga every day.

Worth flagging for anyone running AI in a regulated line of work: Anthropic also reports Opus 5.5 as the strongest-performing model they've tested on their own behavioral audits, with a meaningfully lower rate of boundary circumvention and better resistance to prompt injection.

And, most importantly, there is no 30-day prompt storing like with Fable and Mythos.

In the ongoing battle between OpenAI and Anthropic, Opus 5.5 seems to arrive at the perfect time.

Opus 5.5 is available now in Saga for every plan that already includes model choice.

‍