Anthropic's mid tier Claude line is now the default Claude chatbot for Free and Pro users, and the developer API is priced at $2 per million tokens of input text through Aug.
Anthropic has switched its consumer chatbot and developer API to a new default model, Claude Sonnet 5, and is pricing it at a level that puts near-flagship performance within reach of developers who could not previously afford the company's top tier. The move is a threshold crossing on access, not a research breakthrough, and the price is time-bounded.
The relevant number is $2. That is what Anthropic charges per million input tokens through August 31, 2026, with a matching $10 per million tokens for model-generated output (Anthropic pricing). Free and Pro subscribers on claude.ai are now running Sonnet 5 by default instead of the older Sonnet 4.6 (MacRumors). The temporary price is what makes the launch worth treating as more than a benchmark update: it converts the company's mid-tier line, historically the cheaper, less capable sibling of Opus, into a near-flagship option for indie builders and small product teams.
To understand what changed, it helps to know the tier system. Anthropic sells Claude in two principal lanes. Sonnet is the mid-tier line aimed at consumers and developers who need solid reasoning and coding at a workable price. Opus is the premium line, positioned as the most capable model Anthropic ships at scale. Above both sits Claude Mythos 5, distributed through a restricted program Anthropic calls Project Glasswing; it is not broadly available. Anthropic also markets a variant called Claude Fable 5 as its "most capable widely released" model, distinct from the more restricted Mythos 5.
For most users, what Sonnet 5 actually unlocks is the ability to run multi-step coding jobs and tool-using agents that previously needed either Opus spend or hand-built prompting workarounds. Anthropic describes the model as its "most agentic Sonnet version yet," meaning it plans its own steps, calls external tools such as terminals and browsers, and self-checks output without the user repeatedly nudging it (Anthropic launch post). The official product page says Sonnet 5 can finish complex work that Sonnet 4.6 could not, that it has lower rates of hallucination and sycophancy than earlier Claude versions, and that it sits close to Opus 4.8 across reasoning, coding, knowledge work, and tool use (Anthropic Sonnet page). The model runs with a one-million-token context window and is reachable through the Claude API, Amazon Bedrock, Google Cloud, and Microsoft Foundry, so developers already on those platforms do not need to rebuild their stack. Independent outlets including MarkTechPost, DataCamp, Cosmic JS, and MorphLLM report the same launch facts. Those capability and benchmark claims are company-attributed: Anthropic's own documentation is the primary source in this packet, and no independent benchmark verification is included.
The asymmetry between the temporary price and the permanent price is the part worth flagging. After August 31, 2026, the introductory rate ends and the same API access moves to $3 per million input tokens and $15 per million output tokens, a 50 percent increase on each side (Anthropic pricing). That is still well below Opus pricing, but it reframes the launch as a windowed promotion rather than a permanent price reset. Developers building products on top of Sonnet 5 today should treat the August cliff as a planning milestone.
Opus also does not disappear. Anthropic itself says Opus 4.8 still wins on the hardest agentic tasks (Anthropic Opus page). The honest framing is that Sonnet 5 closes the gap on most everyday reasoning, coding, and tool use at a price most developers can absorb, while Opus remains the right call when a job is genuinely at the frontier. For free and Pro chatbot users, the default switch is effectively a free capability upgrade.
Two broader context lines are worth tracking as the price window plays out. The launch lands in the same news cycle as a reported SEC IPO filing from Anthropic, which raises the question of whether the introductory pricing is partly meant to seed developer adoption ahead of a public-market debut. And Anthropic's top-tier Mythos 5 model, distributed through Project Glasswing, has drawn separate export-control scrutiny, which means the most capable Claude variants are now subject to a different regulatory conversation than the broadly available line.
The thing to watch is August 31, 2026. If the post-intro price holds the same demand pull as the launch, the temporary window did its job, converting a benchmark update into a default-on behavior change for a wide base of users. If the price increase cools adoption, the launch will be remembered more for its introductory pricing than its capability. Either way, that date determines what the release actually meant.