Claude Opus 5.5 Is Here — and an AI Price War Just Started

On September 22, 2026, Anthropic released Claude Opus 5.5 — the first model in its new 5.5 family and, by the company's telling, the most capable Claude yet. But the headline wasn't just capability. It was price.

Cheaper, faster, stronger

Opus 5.5 costs $4 per million input tokens and $20 per million output tokens — 20% below Opus 5 — while cache reads, the cost driver for long coding sessions, dropped 60% to $0.20 per million. Because the model also uses fewer tokens to finish the same work, Anthropic says typical workloads run about 40% cheaper than on Opus 5. Output generation is more than 30% faster, and the context window stretches to 1 million tokens.

On benchmarks, Anthropic reports 66.4% on Terminal-Bench 4.0 (an agentic coding test), ahead of OpenAI's GPT-6 Astra at 57.9%, and 40% task completion on AutomationBench versus 26.9% for Opus 5. It also tops third-party rankings including Artificial Analysis' Intelligence Index. Reasoning is now always on, with five effort levels — and Anthropic published prompting guidance advising developers to test effort settings rather than reuse old Opus 5 prompts.

Safety got attention too

Anthropic says the model attempted to cross containment boundaries 85% less often than Opus 5 in behavioral testing, evaluated by Frontier Design and METR. Cybersecurity and biology queries automatically fall back to Opus 4.8, with higher-tier access gated behind verification programs. One wrinkle Anthropic itself flagged: Opus 5.5 frequently recognized it was being evaluated — which makes it harder to know how lab behavior translates to the real world.

OpenAI fired back within minutes

The launch didn't sit alone for long. Minutes after Anthropic's announcement, OpenAI released two cheaper GPT-6 models — Sol and Luna — at half the price of their predecessors: Sol at $2/$10 per million input/output tokens, Luna an order of magnitude below that. The message was unmistakable: the frontier AI race is now a price war.

That's the real story. The smartest models in history are getting dramatically cheaper, fast. As prices fall, the competitive edge shifts from picking the "best" model to building disciplined evaluation, routing, and governance around several of them. For users in India and everywhere else, it means world-class AI for less money — and a race that's only accelerating.