Anthropic launches Claude 4.8 with reasoning upgrades, undercuts rivals on price

Anthropic on Tuesday announced the general availability of Claude 4.8, the latest version of its flagship large language model, leading public reasoning benchmarks and arriving at roughly half the per-token cost of the previous generation.

The launch — telegraphed for weeks by analysts — intensifies a price war that has reshaped enterprise AI procurement over the past 18 months. OpenAI, Google, and Meta have all slashed pricing on their frontier models in 2026, even as capability gains continue.

What's new

Claude 4.8 introduces:

• A dynamic reasoning mode that allocates compute based on problem difficulty
• Improved tool-use reliability in agentic workflows
• A larger context window of 1 million tokens for enterprise customers
• Native multilingual reasoning that scores higher on Urdu, Arabic, and Bengali benchmarks than any predecessor

The company says the model leads on the AIME 2026 mathematics benchmark and on SWE-bench Verified, an industry standard for measuring autonomous software engineering performance.

Pricing pressure

At $2.50 per million input tokens and $12 per million output tokens, Claude 4.8 lands well below the public pricing of competitors' equivalent tiers. Anthropic CEO Dario Amodei told reporters the company expects "unit economics on inference to keep improving" through the rest of the year.

What it means for developers

For builders in Pakistan and across emerging markets, the price cuts matter as much as the capability gains. Lower per-token costs make it economically viable to deploy AI-powered features in apps targeting low-ARPU users — a market where the previous generation of frontier models simply did not pencil out.

Several Lahore- and Karachi-based startups, contacted for this story, said they were testing the new model in production this week.