How Does Claude Opus 5.5 Perform and Cost?

TL;DR

Claude Opus 5.5 delivers its strongest reported gains in agentic coding and real-world knowledge work while reducing token prices and generating output more than 30% faster than Opus 5. Its best value often appears at medium or high effort, showing that cost per completed task and the chosen thinking setting matter more than token price alone.

Transcript

Opus 5.5 is here. This is the first major model release from Anthropic since they are now pacing the Frontier. So, this is going to be an interesting one. And if you remember, the last time we had a.5 version was late last year when we had Opus 4.5, which really changed the world. That was the inflection point in which coding models became insanely... Read More

Key Insights

• Claude Opus 5.5 is positioned as a frontier model that reaches the level of Fable 5.1 on most tasks while running faster and at a lower cost than Opus 5, with particularly strong reported performance in coding and practical knowledge work.
• Terminal Bench 4.0 is a key measure of agentic coding because it evaluates terminal command execution. Opus 5.5 scored 66.4%, compared with 57.9 for Astra, 55.8 for Fable 5.1, and 52.3 for the previous Opus 5.
• GDPval 2.1 measures real-world knowledge work such as presentation creation, data entry, word processing, and email. Opus 5.5 achieved an 1846 ELO score, compared with 1735 for Fable 5.1 and 1542 for GPT6 Astra.
• Opus 5.5 pricing is $4 per million input tokens and $20 per million output tokens, down from $5 and $25 for Opus 5. Cache reads decrease from 50 cents to 20 cents, while cache writes decline from $6.25 to $5.
• Cost per completed task is more informative than token price alone because an inexpensive model can still cost more if it requires many additional tokens to reach a result. Opus 5.5 combines lower prices with higher benchmark quality and more than 30% faster output generation.
• Thinking effort strongly affects both quality and expense. On Frontier Code, the medium-effort configuration reportedly scored higher while costing under $1 per task than the maximum-effort configuration, which cost more than $5 per task, showing that additional computation does not guarantee better results.
• Opus 5.5 does not lead every reported benchmark. Astra scored 41.4 on Automation Bench versus 40 for Opus 5.5, while GPT6 Astra led Terminal Bench Science, although Opus 5.5 still improved over Fable 5.1 on that science evaluation.
• Opus 5.5 reportedly communicates more naturally than Opus 5 by placing important information first and following supplied writing rules. The reviewer views it primarily as a coding and knowledge-work model, while expressing uncertainty about how it compares with Astra for creative writing.
• More videos from this channel:
  • How to Cut AI Costs With a Multi-Model Workflow
  • How does Deepseek V4.1 Flash improve efficiency and open weights perf…
  • How to test VO3 AI video generation capabilities
• More from this channel: Matthew Berman

Questions & Answers

Q: How well does Claude Opus 5.5 perform on coding benchmarks?

Claude Opus 5.5 shows its clearest reported advantage on agentic coding evaluations. It scored 66.4% on Terminal Bench 4.0, ahead of Astra at 57.9, Fable 5.1 at 55.8, and Opus 5 at 52.3. It also scored 54.4 on Frontier Code V1.1 and 57.8 on Cursorbench, indicating sizable improvements over several cited comparisons.

Q: What does Claude Opus 5.5 cost per million tokens?

Claude Opus 5.5 costs $4 per million input tokens and $20 per million output tokens. The previous Opus 5 prices were $5 for input and $25 for output, making the listed reduction 20%. Cache reads fall from 50 cents to 20 cents, and cache writes decline from $6.25 to $5.

Q: Why is cost per task more useful than token pricing?

Cost per completed task accounts for both token price and the number of tokens required to produce a successful result. A model with inexpensive tokens can remain costly if it uses many more of them to solve the same problem. The analysis therefore combines benchmark quality with estimated task cost to identify configurations that deliver strong results efficiently.

Q: Which Claude Opus 5.5 effort setting offers the best value?

The reported charts suggest that medium or high effort often provides the strongest balance of quality and cost, although the best choice depends on the task. On Frontier Code, medium effort reportedly cost under $1 per task and scored higher than maximum effort, which exceeded $5. On Terminal Bench, medium and high settings also appeared especially competitive.

Q: How does Claude Opus 5.5 perform on knowledge work?

Claude Opus 5.5 achieved an 1846 ELO score on GDPval 2.1, compared with 1735 for Fable 5.1 and 1542 for GPT6 Astra. GDPval is described as testing real-world knowledge work, including presentation creation, data entry, word processing, and email. The result is presented as one of the model's most important improvements.

Q: Is Claude Opus 5.5 the best model on every benchmark?

Claude Opus 5.5 does not rank first on every cited evaluation. Astra scored 41.4 on Automation Bench compared with 40 for Opus 5.5. GPT6 Astra also placed ahead on Terminal Bench Science, although Opus 5.5 scored 58 and improved over Fable 5.1. Its visual chart recognition result, 89%, was also close to Fable 5.1 at 88.4.

Q: How much faster is Claude Opus 5.5 than Opus 5?

Claude Opus 5.5 is reported to generate output more than 30% faster than Opus 5. The speed increase accompanies lower listed input and output token prices. According to the analysis, this combination matters because faster generation and efficient reasoning can reduce the total cost and waiting time associated with completing practical coding or knowledge-work tasks.

Q: Is Claude Opus 5.5 suitable for creative writing?

The available analysis does not establish Claude Opus 5.5 as the leading creative-writing model. The reviewer considers Astra the strongest current option for creative prose because it has less of an artificial style, while expressing interest in testing Opus 5.5 further. Based on the presented evidence, Opus 5.5 is positioned mainly for coding and knowledge work.

Summary & Key Takeaways

• Claude Opus 5.5 is presented as a faster and cheaper frontier model focused primarily on coding and knowledge work. Its highlighted results include 66.4% on Terminal Bench 4.0, 54.4 on Frontier Code V1.1, and an 1846 ELO score on GDPval 2.1, substantially exceeding the cited comparison results.

• Pricing falls from $5 to $4 per million input tokens and from $25 to $20 per million output tokens compared with Opus 5. Cache reads cost 20 cents instead of 50 cents, while cache writes fall from $6.25 to $5. Output generation is also reported as more than 30% faster.

• The analysis emphasizes cost per completed task rather than price per token. Benchmark charts suggest that medium or high effort can provide better value than maximum effort, which sometimes costs much more without producing a better score. The model also reportedly communicates more naturally and places important information first.