Anthropic launches Claude Sonnet 5: the affordable AI agent that can actually finish the job

Anthropic’s new Claude Sonnet 5, released 30 June 2026, is the first model in the Sonnet line that can reliably finish a complex multi-step task without stalling halfway through.

That one capability change is what separates it from its predecessors, and it matters directly to any Kenyan business that has tried to build an AI agent and found it giving up midway through a workflow. Sonnet 5 is available now as the default model on Claude’s free and Pro plans, and through the API at $2 per million input tokens and $10 per million output tokens through 31 August 2026.

What is actually new

The benchmark numbers tell part of the story. On agentic coding, Sonnet 5 scores 63.2 percent, up from Sonnet 4.6’s 58.1 percent and closing to within 6 points of Opus 4.8’s 69.2 percent. On knowledge work, it outperforms Opus 4.8. Those numbers are meaningful but not the main event.

The main event is the behavioral change. Sonnet 5 can make plans, use tools including browsers and terminals, and run autonomously through workflows that would have required a more expensive model to complete reliably. Zapier engineer Daniel Shepard, quoted by TechCrunch, described it directly: “That used to stall halfway. For day-to-day automation, it’s a no-brainer.”

Anthropic also says Sonnet 5 checks its own output without being prompted to do so, hallucinates at a lower rate than Sonnet 4.6, and is harder to manipulate through prompt injection, meaning adversarial text embedded in data it processes is less likely to hijack the agent’s behavior.

The pricing reality for Kenyan businesses

At $2 per million input tokens and $10 per million output tokens, Sonnet 5 costs roughly one-fifth of what Opus 4.8 costs per call. Converting to Kenyan shillings at the current exchange rate of approximately KSH 130 per dollar:

• Processing 1,000 loan applications with a credit-scoring agent: roughly KSH 260 in input costs plus output costs
• Running a WhatsApp automation that handles 500 customer queries per day: under KSH 20 per day
• Generating a daily market intelligence brief for an agribusiness exporter: under KSH 5

These are not hypothetical savings. They are the difference between AI automation that is financially viable for a Nairobi SME and AI automation that only makes sense for a large corporation with a dedicated AI budget.

What this means for Kenyan software developers and AI builders

The East African developer community has been building on Claude’s API for workflows ranging from M-Pesa transaction analysis to drought risk assessment for insurers. The consistent limitation has been that the models capable of finishing complex multi-step reasoning cost too much to run at scale on African market margins.

Sonnet 5 moves the frontier. A model that can browse the web, reason about what it finds, call an external API, check its own output, and return a synthesised result without human intervention in the middle was, until this week, an Opus-tier capability at an Opus-tier price. It is now a Sonnet-tier price.

The price is discounted through 31 August 2026. After that, it moves to $3 per million input tokens and $15 per million output tokens. For businesses with ongoing AI agent infrastructure, building at the current rate and understanding the revised cost structure before September matters.

An honest limitation

Sonnet 5 is better than Sonnet 4.6, substantially so on agentic tasks. It is not as accurate as Opus 4.8 on the most demanding reasoning tasks. For workflows where a single wrong inference has a significant consequence, such as medical flagging, legal document review, or high-value lending decisions, Opus 4.8 remains the appropriate choice. For the broad category of business automation, research workflows, and customer-facing agents, Sonnet 5 is the right model at the right price.

If you are ready to build AI agent workflows for your business and want to understand what Sonnet 5 can do for your specific operations, WhatsApp us on 0711 344 702. The cost structure has changed enough that projects that were not viable three months ago now are.