Claude Sonnet 5, released by Anthropic, ranks fifth in the Artificial Analysis Intelligence Index with 53 points, tying with GPT-5.5 (high). The model outperforms the pricier Opus 4.8 on some agent-based tasks, but its increased token consumption makes it more expensive per task. According to Artificial Analysis, an average task with Sonnet 5 costs $2.29, compared to $1.97 with Opus 4.8. This increase is due to Sonnet 5 using about 40 percent more output tokens per task at maximum performance settings. | Image: Artificial Analysis
Anthropic maintains the same token prices for Sonnet 5 as its predecessor: $3 per million input tokens and $15 per million output tokens. However, the model's higher token usage leads to a nearly doubled cost per task, even though it performs better than Opus 4.8 on some benchmarks. The company is currently offering a promotional rate of $2 or $10 per million tokens through September 1, but Artificial Analysis based its findings on regular prices. This pattern of hidden cost increases mirrors past practices with models like Opus 4.7, where token counts were inflated without changing the official pricing. | Image: Artificial Analysis
Anthropic's models have consistently seen rising costs without clear price adjustments. When Opus 4.7 launched, token prices remained the same, but a new tokenizer increased token counts by about 30%, leading to higher real costs. With Sonnet 5, the model's more agentic behavior further compounds this issue, consuming far more tokens per task. As a result, Anthropic's models become pricier with each generation, yet official price lists do not reflect these changes. This lack of transparency contrasts with Chinese competitors like Deepseek V4 Pro and GLM-5.2, which offer similar performance at lower costs. | Image: Artificial Analysis
Source: thedecoder
