Anthropic Claude 3.7 Sonnet
Anthropic billed Claude 3.7 Sonnet as its most intelligent model to date and 'the first hybrid reasoning model on the market' — a single model that answers near-instantly or, on demand, thinks step-by-step with a developer-controllable thinking budget of up to 128K output tokens. It claimed standout gains in coding and front-end web development, with state-of-the-art results on SWE-bench Verified (about 70.3% at high compute). Alongside the model it unveiled Claude Code, its first agentic command-line coding tool, as a limited research preview.
It shipped the same day across all Claude plans (Free, Pro, Team, Enterprise), the Claude Developer Platform, Amazon Bedrock, and Google Cloud Vertex AI, with extended thinking on every paid tier; pricing held steady at $3/$15 per million input/output tokens (thinking tokens billed as output). Claude Code launched concurrently as an invite-limited research preview.
Claude 3.7 Sonnet folded a reasoning mode into a mainstream model without forcing a separate 'thinking' SKU, an approach the rest of the industry effectively converged on. Its real significance was less the benchmark and more what it catalyzed: paired with Claude Code, it anchored the 2025 agentic-coding wave and made Anthropic the default for AI-assisted software development, fueling fast-growing API revenue. Observed limitations were real — extended thinking raised latency and token cost, and some users found it over-eager to make code changes — but developer adoption was strong and sticky. Holding price flat while adding reasoning was a deliberate, aggressive competitive move.
It kicked off the agentic-coding era and made Anthropic the go-to vendor for AI software development — turning 'AI writes and runs code in your repo' from demo into a daily workflow executives now budget for.