Breaking

Anthropic's Claude Opus 5.5 Challenges GPT-6 Astra on Agent Benchmarks at Lower API Prices

September 22, 2026 at 19:16 EDT

Foundation Models
AI Agents
Software Dev & Coding

Anthropic has put Claude Opus 5.5 into the center of the frontier-model race, positioning the new Opus release against OpenAI's GPT-6 Astra on agentic coding, professional work and cost per task. The launch sharpens a comparison that developers are already making: Astra still posts the strongest public numbers in some science and automation tests, while Opus 5.5 is being framed as a cheaper, coding-heavy alternative for long-running agent workflows.

Sep 22, 2026 · Anthropic

Claude Opus 5.5 Takes Aim at GPT-6 Astra

Anthropic's new Opus model competes with Astra on agentic coding benchmarks while listing lower API token prices. The split is not absolute: Astra still leads selected automation and science tests.

Input price comparison

$4Claude Opus 5.5
per 1M input tokens

$10GPT-6 Astra
per 1M input tokens

Astra's listed input price is 2.5x Opus 5.5's listed input price.

66.4%Opus 5.5 on Terminal-Bench 4.0

57.9%GPT-6 Astra on Terminal-Bench 4.0

41.4%Astra on AutomationBench

Why developers like it

Lower token prices and cache-read economics make Opus 5.5 attractive for long coding agents, code audits and repeated tool loops.

What tempers the result

Benchmark effort settings differ, some margins are narrow, and always-on thinking means real task cost still depends on output volume.

The buyer takeaway

Coding agents→Measure task cost→Route by workload

Opus 5.5 looks strongest where agent loops and price matter. Astra remains competitive where broader science, automation and computer-use performance dominate.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.