Anthropic has released Claude Sonnet 5.5, a new version of its mid-tier model that promises faster performance and lower costs, ultimately designed to allow teams to spawn multiple AI agents without exceeding cost limits.
The new model, which Anthropic has described as having a “sharp eye for design”, could see high engagement with the creative industry in Australia, with creative and marketing teams among those already experimenting heavily with agentic AI workflows.
Anthropic has described Sonnet 5.5 as a “clear upgrade over Claude Sonnet 5”, claiming it generates outputs more than 30 per cent faster and typically requires significantly fewer tokens to complete the same work.
Sonnet 5.5 is priced the same as its predecessor at US$2 per million input tokens (around A$3), US$10 per million output tokens (around A$15), and US$0.20 per million tokens for cache reads. Put simply – tokens are small chunks of text processed by the AI. ‘Input tokens’ are what you send in, while ‘output tokens’ are what the model generates back.
However, Anthropic says the model can cost up to 30 per cent less per task because it uses fewer tokens to complete the same work.
The announcement comes at a time when there has been increasing industry discussion around “token-maxxing”, token burnout and the escalating cost of running increasingly sophisticated AI agents.
In a recent interview with ACAM, the Australian Council of Advertising and Marketing, the organisation warned marketers against what it described as “jackhammer AI syndrome” — using powerful, expensive AI models for simple tasks — while also cautioning that brands are entering “arranged marriages” with enterprise AI platforms that may not actually suit their needs.
And ACAM’s 2026 Australian AI in Marketing Benchmark Report found 83 per cent of organisations remain stuck in the early stages of AI maturity, with no organisation reaching the top two levels of capability.
The report, based on responses from 126 Australian CMOs and senior marketing leaders across 12 industries, found almost six in 10 organisations (58 per cent) have no documented AI roadmap, while 45 per cent said their enterprise AI platforms were disappointing, harder than expected or only “OK”.
‘Time will only tell’
B&T sat down with ACAM co-founder Kathryn Illy on Tuesday morning to discuss what the arrival of Sonnet 5.5 could mean for Australia’s creative industry — and whether making agentic AI cheaper will actually solve the problems marketers and agencies are already facing.
Illy, who brings more than 20 years of experience across B2B and B2C marketing, said the new model is likely to see high adoption across the creative industry, with creative agencies and marketing teams in particular likely to engage with it.
While it is still early days, she said “creative agencies are likely to be the first to jump on this model”, particularly as the lower inference costs make it easier to experiment with AI agents at scale.
The attraction, she said, is not simply the model itself, but what happens when teams can afford to run more AI-powered workflows without the same concern around cost.
For Illy, however, the bigger question is what agencies and marketers actually do with the additional capacity.
“Time will only tell,” she said.
Could the model just result in ‘slop grenades’ and ‘more chaos?’
Illy believes the updated model will create risks within the industry as well as opportunities.
While it will likely “deliver strong results and reduce costs”, she said the technology could also amplify existing problems if organisations have not put the right structures around AI deployment.
ACAM’s recent CMO report, for example, found “big gaps” around team design, with Illy pointing to workflow as a major structural gap for Australian marketing teams.
The report found many organisations are still lacking a documented roadmap or clear team design around how AI should be deployed, with questions around capability, governance and accountability still unresolved.
Illy said this becomes “increasingly important” as models become cheaper and easier to deploy.
“If you don’t have those basic principles in place, or you don’t have the team or capability, or appropriate governance, then this model is going to create even more chaos,” she told B&T.
“Yes, this is commercially attractive, but if you don’t have those organs, and those basic principles and appropriate governance in place it’s just going to create even more chaos”.
“Having those foundations in place will become critically important.”
She added that cheaper and more capable AI “does not necessarily mean better AI deployment”. Instead, Illy said organisations could end up producing more of what they are already producing — only faster.
That means the industry could see a wave of more AI-generated content, more automated workflows and more agents operating across marketing functions.
This creates a risk, she suggested, of creating “slop grenades” (the latest term of AI slop) – giving teams more capacity to produce work without first establishing where AI should, and should not, be used.
Illy said those already using AI within their creative teams or marketing workflows are likely to see those existing applications amplified as agentic models become cheaper and easier to deploy.
“We’re likely to see those areas where it’s being used amplified even more,” she said.
That includes “areas such as content generation, brand management, customer journeys and measurement”.
“We know marketing teams are busy, so these are the key areas that marketing teams are focusing on with deploying AI agents and automation,” she said.
‘A staple’ for agencies
And does the creative industry back it?
B&T spoke with Joshua Flowers, co-founder and chief commercial officer of Sydney creative agency Paper Moose, who admitted his own personal AI token usage over the past 30 days amounted to US$15,845 (A$22,601.65).
Flowers said the figure demonstrates just how quickly inference costs can add up once AI agents are being used intensively.
“Agentic workflows are already hugely powerful if wielded by skilled AI wranglers,” Flowers told B&T.
But as those workflows expand in usefulness, so too does the cost of running them.
Flowers said that while users on subsidised AI subscriptions can access significant amounts of inference for relatively small monthly fees, the underlying API costs can quickly balloon.
For Flowers, that is what makes the arrival of Sonnet 5.5 particularly interesting.
“It’s the solid mid-tier option designed to make both the inference cost less on the AI lab’s side, which they desperately need for profitability, but also to be more affordable for the end user,” he said.
And the agency use cases are already moving well beyond simple copy generation.
Aside from coding, Flowers said emerging applications for agencies include graphic design layouts in Figma, motion design, video editing, game design, production and increasingly basic 3D modelling.
“We do this internally already for asset layouts,” he said of using AI for graphic design layouts in Figma.
“Motion design — Opus 5.5 and Sonnet 5.5 are shockingly good at creating motion design videos from a single prompt end to end.”
“Claude Code, Codex, Opencode etc alongside models like Sonnet 5.5 will obviously become a staple set of tools for agencies,” Flowers said.
He expects the biggest impacts to continue across “coding, graphic design, motion design, video editing, game design and production and increasingly some basic 3D modelling as well”.
But Flowers believes the lower cost of AI will not necessarily mean agencies spend less on AI, he believes the usage to expand.
As AI becomes cheaper and more efficient, he pointed to Jevons’ paradox — the idea that when a technology or resource becomes more efficient or cheaper to use, overall consumption can increase rather than decrease.
“As Jevon’s paradox shows, when a technology or resource becomes cheaper or more efficient to use, overall demand expands so much that total consumption goes up rather than down,” he said. “So I expect Sonnet 5.5 to continue to drive mainstream adoption of these powerful agentic tools.”
B&T has approached Anthropic for comment.
