Anthropic's Claude Sonnet 5 cuts the cost of running AI agents
Anthropic's new midsize Sonnet 5 undercuts Opus and rival flagships on price while keeping agentic muscle. For India's automation builders, the maths just changed.
The News
Anthropic has released Claude Sonnet 5, a midsize model the company is pitching squarely at businesses that want to run autonomous software agents without paying flagship prices. The model went live on Tuesday, 30 June 2026, and immediately became the default option for both the free and Pro subscription tiers.
The headline is the bill. Through 31 August, Sonnet 5 costs 2 dollars per million input tokens and 10 dollars per million output tokens. After that promotional window closes, input pricing rises to 3 dollars per million tokens while output stays at 10. That places it below Opus 4.8, OpenAI's GPT-5.5 and Google's Gemini 3.1 Pro, though it remains dearer than the budget-tier Gemini 3.5 Flash.
On capability, Anthropic says the model can make plans and use tools such as browsers and terminals to complete multi-step jobs on its own. It scored 63.2 percent on an agentic coding benchmark, trailing the larger Opus 4.8 at 69.2 percent but well ahead of the previous Sonnet 4.6 at 58.1 percent. On general knowledge work, Anthropic claims Sonnet 5 edges past even Opus 4.8.
Why It Matters
The pricing tells you where the contest has moved. A year ago the bragging rights were about raw intelligence and benchmark crowns. Now the fight is over cost per completed task, because agents that loop through dozens of tool calls burn tokens fast, and a few dollars per million either makes an automation viable or kills it. Sonnet 5 lands the same week that OpenAI's GPT-5.6 Sol and Google's Gemini 3.5 Flash chase the identical promise: capable autonomy at a price that survives contact with a finance team.
This is the commoditisation pattern that followed every earlier capability jump. When GPT-4 arrived in March 2023 it was a luxury good; within eighteen months equally capable models cost a fraction of the price. The midsize tier is where that compression bites hardest, and Anthropic is signalling it would rather own the high-volume workhorse slot than defend a premium it cannot hold. Daniel Shepard of automation platform Zapier, an early tester, said tasks that "used to stall halfway" now run to completion.
Indian Angle
For India's services and product economy, cheaper agentic models are an input cost, not a novelty. The country's IT majors and the growing band of agent-focused startups bill clients on margins sensitive to every dollar of inference. A model that does most of Opus-grade work at a third of the headline cost directly improves the unit economics of code-generation, support-automation and back-office agents being sold out of Bengaluru, Pune and Hyderabad.
It also sharpens the question facing home-grown model builders such as Sarvam and Ola's Krutrim. Their pitch has leaned on sovereignty, Indian-language strength and data residency rather than price. With a global lab now offering 2 dollars per million input tokens, the domestic case has to rest more firmly on those differentiators, because raw cost is no longer a moat anyone can defend for long.
For India's enterprises and regulators, the default-model switch matters too. When millions of free and Pro users are quietly moved onto an agentic model that drives browsers and terminals, MeitY's evolving guidance on autonomous systems becomes less theoretical and more a near-term compliance question for any Indian firm wiring these agents into live workflows.
FAQ
When does the launch pricing end?
The promotional rate of 2 dollars per million input tokens runs through 31 August 2026. After that, input pricing moves to 3 dollars per million tokens, while output pricing stays at 10 dollars per million throughout.
How does it compare to Opus 4.8?
On agentic coding Sonnet 5 scored 63.2 percent against Opus 4.8's 69.2 percent, so the larger model still leads on complex coding. But Anthropic says Sonnet 5 slightly outperforms Opus 4.8 on general knowledge work, at a far lower price.
What does this mean for Indian developers?
Lower input and output costs cut the inference bill for token-hungry agents, improving margins for Indian startups and IT firms that build and resell automation. It also raises the competitive bar for domestic models like Sarvam and Krutrim.
Where can I read the original announcement?
TechCrunch reported the launch in detail, including pricing tiers and benchmark figures. The link is in the attribution note below.
This story was reported by TechCrunch. Read the full original coverage at TechCrunch.