"Anthropic has launched Claude Opus 4.5 with a 67% price cut that repositions its flagship model from a boutique offering to a production-ready enterprise tool. The new pricing of $5 per million input tokens and $25 per million output tokens - down from $15 and $75 - brings Anthropic closer to OpenAI and Google while maintaining a premium position. The launch comes a week after Google released Gemini 3 and less than two weeks after OpenAI launched GPT-5.1, underscoring the rapid pace of competition in enterprise AI."
"Anthropic claimed Opus 4.5 achieved 80.9% on Software Engineering Benchmark Verified, outperforming OpenAI's GPT-5.1-Codex-Max at 77.9%, Google's Gemini 3 Pro at 76.2%, and its own Sonnet 4.5 at 77.2%. The company also said the model scored higher on its internal two-hour performance engineering assessment than any human candidate who has taken the exam."
""Benchmark scores often look impressive but mean very little once a model enters production," said Sanchit Vir Gogia, chief analyst at Greyhound Research. "They are clean, simple, and run in isolation. Most enterprise systems are layered with legacy software, inconsistent workflows, and regulatory overhead." Leslie Joseph, principal analyst at Forrester, noted that while Anthropic's lead is statistically significant, "the gap between these models has narrowed to the point where winner-takes-all metrics are less relevant than considerations of best architectur"
Claude Opus 4.5 arrives with a 67% price reduction to $5 per million input tokens and $25 per million output tokens, down from $15 and $75, positioning the model for broader enterprise adoption. Pricing now approaches OpenAI and Google while remaining premium versus competitors like GPT-5.1 and Gemini 3 Pro. Opus 4.5 posted 80.9% on the Software Engineering Benchmark Verified and reportedly outscored humans on an internal two-hour performance engineering exam. Benchmark advantages exist, but production realities, integration complexity, regulatory overhead, and narrowing gaps between top models shift focus toward architecture, integration, and compliance.
Read at Computerworld
Unable to calculate read time
Collection
[
|
...
]