Opus 5.5
Anthropic's flagship large language model, built on a Mixture-of-Experts (MoE) architecture. It is designed for high-performance reasoning and multimodal tasks, and is positioned as a competitive alternative to top-tier models at a lower price point.
Core Facts
Google stated that after the introductory period ends, the price of Gemini 4 Argon will rise to $4 per million input tokens and $20 per million output tokens, positioning it at the Opus 5.5 level.
Opus 5.5 reaches the level of Fable 5.1 on most tasks, but with approximately 40% lower operating costs and faster speed.
Longer inference time does not always yield better results; models may fall into the trap of overthinking, and what truly determines output quality is the model's foundational capabilities and degree of alignment.
Timeline (last 90 days)
Muse Wonder costs $4.52 per task call, an increase of nearly 3 times compared with Muse Spark 1.3, about $1 more expensive than GPT-6 Astra or Grok 4.7, but about $1.5 cheaper than Opus 5.5.
Developer Theo placed WonderFull in his S-tier models, tied with Opus 5.5.
AI researcher Mehu Mohan said he was also dissatisfied with WonderFull, believing it is far inferior to Opus 5.5 and even worse than GPT-6.1 Sol.
Gemini 4 Argon scored 52.8% on CursorBench at default reasoning effort, putting it on the same level as Opus 5.5.
In the DeepSWE 1.1 benchmark, Muse Wonder Full achieved 79.0%, exceeding Claude Opus 5.5's 74.2% and GPT-6 Astra's 74.1%.
In the same game at the same token price, Opus 5.5 completed it in 271 turns, costing $8.44.
On MemoryBench, Wonder defeated Fable, Opus, and Astra; on the Artificial Analysis index, Wonder Full ranks second, behind only Opus 5.5.
Tim Sweeney said the capability Opus 5.5 shows with very few instructions strikes him as quite crazy.
@lnkiai's experiment shows: faced with the same "make a 3D avatar" task, Codex (GPT-6.1 Sol) chose to model from scratch in Blender, while Claude Code (Opus 5.5) proposed finding distributed models and parts and then assembling and machining them.
@aicreataro's cross-model creative pipeline stack is: characters from MiniMax H3, key visuals from Midjourney, music from MiniMax Music 3.0, and Opus 5.5 driving Blender via Python.
40 more timeline events
All Facts (20)
Google stated that after the introductory period ends, the price of Gemini 4 Argon will rise to $4 per million input tokens and $20 per million output tokens, positioning it at the Opus 5.5 level.
80%VerifiedOpus 5.5 reaches the level of Fable 5.1 on most tasks, but with approximately 40% lower operating costs and faster speed.
80%VerifiedLonger inference time does not always yield better results; models may fall into the trap of overthinking, and what truly determines output quality is the model's foundational capabilities and degree of alignment.
75%VerifiedOn Automation Bench, Gemini 4 Argon scored 51.3%, surpassing Fable 5.1's 31.4% and Opus 5.5's 42.5%.
70%VerifiedOpus 5.5 scored 66 on the Terminal agent benchmark, higher than Sonnet's 55 and Astra's 57.9.
70%VerifiedAnthropic states that Opus 5.5 can reduce total costs for typical tasks by approximately 40% and improve output speed by more than 30%.
70%VerifiedClaude Opus 5.5 generates output more than 30% faster than Opus 5.
70%VerifiedThe company states that Sonnet 5 shows significant improvements over Sonnet 4.6 in reasoning, tool use, coding, and knowledge work, with overall performance approaching OPUS 4.8 but at a lower price.
70%VerifiedOn the SWE Bench Pro programming benchmark, Sonnet 5 achieved 63.2% and Opus 4.8 achieved 69.2%.
70%VerifiedMistral Large 4 narrowly outperformed GPT-6 Astra on visual grounding tasks with a score of 42% versus 41%.
65%VerifiedHaiku 5.5 can work as a sub-agent alongside Claude Opus 5.5 and Sonnet 5.5 to complete programming tasks.
65%VerifiedIn the deep audit benchmark, Sonnet 5.5's cost is approximately half that of Opus, its score is slightly higher than Opus, and it takes about 5 minutes (roughly half the time Opus takes).
65%VerifiedGPT-6 Astra is notably different from previous model iterations, being powerful in certain respects while showing little improvement or even appearing to regress in other scenarios.
65%VerifiedThe cache read mechanism retains the KV cache for recurring long prompt prefixes, charging at an extremely low rate upon a cache hit; the leaked cache read price for Opus 5.5 is approximately 1/20 of the input price.
65%VerifiedThe core challenges of long-running Agents include tasks easily drifting from goals, errors cascading and amplifying, and resource consumption being difficult to estimate.
65%VerifiedOpus 5's pass rate at Low Effort is approximately 60%, surpassing Opus 4.8's 59% at the Max level, for which the per-task cost is as high as $13.
65%VerifiedOn the SWE-Bench Pro benchmark, Sonnet 5 significantly trails Opus 4.8.
65%VerifiedClaude Opus 5's capabilities are close to the level of the previous-generation flagship Opus, but with significantly reduced inference costs.
65%VerifiedGPT-5.6 Sol has the capability to handle long-cycle tasks running continuously for days, requiring only a given objective to iteratively produce detailed results, with significantly improved UI generation quality.
65%VerifiedAnthropic released Claude Sonnet 5, officially claiming its performance approaches that of Opus 4.8 at a lower price
60%