Anthropic announced its latest large language model, Claude Opus 5.5, on September 22, 2026. Succeeding Opus 5, which launched in July, the new model cuts usage prices by as much as 40 percent while making responses more than 30 percent faster. Across several benchmarks focused on coding and knowledge work, Opus 5.5 outperforms its predecessor.

Pricing cut across the board

Input token pricing for Opus 5.5 drops to 4 USD per million tokens, down 20 percent from Opus 5's 5 USD. Output tokens now cost 20 USD per million tokens, also a 20 percent reduction from 25 USD. Cache-read pricing falls even further, to 0.20 USD per million tokens, a 60 percent cut from 0.50 USD. Anthropic says that for typical workloads reflecting real-world usage patterns, overall costs drop by roughly 40 percent on average.

※1 USD = 155 JPY (based on mid-September 2026 market rates)

The price cuts stem largely from more efficient use of compute. Anthropic attributes the pricing change directly to achieving the same performance with less computational cost.

Stronger results in coding and knowledge work

On the benchmark side, Opus 5.5 scored 66.4 percent on Terminal-Bench 4.0, a software engineering evaluation, and 54.4 percent on FrontierCode v1.1, which measures coding ability. In one real-world example, the model completed a code migration spanning 680,000 lines in under a day.

For knowledge work, Opus 5.5 earned an Elo score of 1,846 on GDPval-AA v2.1, an evaluation covering 44 professions. Anthropic describes this as the strongest performance it has measured to date for any of its models. Anthropic also says Opus 5.5 performs at the level of its larger Claude Fable 5.1 model on most work, with particular strength in agentic coding, computer use, and knowledge work.

Faster responses, stronger safety

Response generation is more than 30 percent faster than Opus 5. Anthropic also refined how the model communicates, placing key information earlier in responses and reducing unnecessary jargon.

On the safety front, Opus 5.5 achieved the strongest results Anthropic has recorded in automated behavioral audits, and its resistance to prompt injection attacks has also improved. As model capabilities continue to advance, Anthropic appears to be pairing that progress with parallel safety work.

Availability and what comes next

Claude Opus 5.5 became available the same day it was announced, through the Claude Platform as well as major cloud providers including Amazon Web Services, Google Cloud, and Microsoft Azure. Usage limits across plans have also been raised.

Anthropic says the mid-tier Claude Sonnet 5.5 and the lightweight Claude Haiku 5.5 will follow with similar improvements within the coming weeks, keeping the three-tier Opus, Sonnet, and Haiku lineup intact.

Summary

Claude Opus 5.5 cuts prices by as much as 40 percent while improving coding and knowledge-work benchmark scores and speeding up responses by more than 30 percent, alongside stronger safety results. Anthropic plans to roll out Sonnet 5.5 and Haiku 5.5 within the coming weeks, marking a notable move in both performance and cost as AI development competition continues.

Sources: [1] https://www.anthropic.com/claude-opus-5-5