The new midsize model, priced at $2 per million input tokens through August, performs near the flagship Claude Opus 4.8 on multi step coding and knowledge work, undercutting OpenAI's GPT 5.5 and Google's Gemini 3.1 Pro on cost per task.
Anthropic's Claude Sonnet 5 sits at a price point that makes mass deployment of agentic AI economically viable in a way the company's previous lineup never did. Agentic AI is the industry's term for autonomous workflows that plan, write code, run tests, and revise across multiple steps without a human in the loop. The midsize model launched Tuesday at $2 per million input tokens and $10 per million output tokens through August 31, 2026, then steps up to $3 per million input tokens afterward, according to TechCrunch's launch report. That is meaningfully cheaper than Anthropic's flagship Claude Opus 4.8, OpenAI's GPT-5.5, and Google's Gemini 3.1 Pro, though it remains more expensive than Google's Gemini 3.5 Flash.
For builders running agents in production, the math changes shape at this kind of price. An agent that loops through dozens of model calls to plan, write code, run tests, and revise will now spend a fraction of what the same loop would have cost on Opus. The launch puts Sonnet 5 inside the free and Pro subscription tiers and makes it the default Anthropic model for every plan, per TechCrunch, which means the cost story is not just about developer API bills; it is about what every Anthropic customer is now pointed at by default.
Anthropic's own Claude pricing page and models overview are the authoritative references for current list prices and tier placement. The pricing numbers above should be rechecked against those pages before any production commitment is built on them.
The benchmark story is mixed but tilted toward Sonnet 5's favor for most production workloads. On an agentic coding evaluation that requires the model to plan, execute, and debug across multiple steps, Sonnet 5 scores 63.2 percent, ahead of the prior Sonnet 4.6 at 58.1 percent but behind Opus 4.8 at 69.2 percent, per TechCrunch's launch reporting. On a separate knowledge-work benchmark, Sonnet 5 slightly outperforms Opus 4.8. That is the pricing-to-performance crossover builders will care about: Sonnet 5 does not beat Opus on the hardest coding work, but for the bulk of agentic office tasks it is close enough that the price gap starts to dominate the decision.
The competitive picture is what turns this from a product launch into an industry signal. OpenAI shipped its most agentic model, GPT-5.6 Sol, in preview the prior week, with subagent task splitting as the headline pitch. Google launched Gemini 3.5 Flash in May, framed as an agentic tool for planning, building, and iteration, and undercut Anthropic on price. Anthropic is now responding in the same lane, with a tier-above-flash model that is also priced aggressively, according to TechCrunch's coverage.
That is the part to watch. Each of the three labs has now published a model aimed at the same buyer: a team that wants to ship autonomous workflows without paying flagship-model rates. The competitive floor on agent pricing is dropping faster than the ceiling, which means the question for buyers is no longer whether agents are affordable but which provider's tool-calling, context, and reliability story holds up at the new price point.
Two caveats worth keeping in front of the launch. First, the $2-per-million input price is a launch promotion through August 31, 2026, and reverts to $3 afterward, per TechCrunch. Production budgets built around the launch price should be stress-tested against the post-August rate. Second, the benchmarks cited above are the vendor-adjacent numbers TechCrunch reported from Anthropic's launch material; independent reproduction on real agent workloads is the variable that will decide whether the unit-economics promise holds up.
The next trigger worth tracking is how OpenAI prices GPT-5.6 Sol out of preview and whether Google widens the Gemini 3.5 Flash feature gap. The agent-economics story only matters if all three labs keep pulling the floor down together; if any of them steps back, the others inherit pricing power they do not currently have.