Search results for Claude Haiku 4.5
AI & Enterprise
Cheaper token price can raise total coding AI cost by 14 times, study finds
An experiment found that in coding tasks an AI model with a lower token price can end up costing more to fix bugs. A Fiddler research team compared Anthropic\'s Claude Haiku 4.5 and Claude Sonnet 5 on five Python bug-fix tasks. Across 45 runs per model, Sonnet cost $0.45 while Haiku cost $4. Haiku generated much longer outputs, including results for commands not executed, raising both output and subsequent input token costs.
AI & Enterprise
Upstage unveils \'Solar Open 2\' touting agent performance
Upstage unveiled its open-weight large language model \'Solar Open 2\' optimised for agent task execution and released the model weights on Hugging Face. Developed through a government sovereign AI foundation model project, it uses a mixture-of-experts design with 250 billion parameters and activates 15 billion in use. Upstage said it outperformed rival models on multiple benchmarks and plans broader deployments and AX services across industries.
AI & Enterprise
GitHub Copilot fully implements token-based billing as users complain monthly limits shrink fast
GitHub Copilot users are voicing growing complaints about a new pay-as-you-go billing system based on AI credits. Reports say monthly limits can be depleted quickly even with typical use. Plans now include fixed monthly credit allocations, with credits deducted by actual usage and model pricing. Some users say converted costs would rise to thousands of dollars. Others say changing usage habits can reduce spending, while model-to-model cost gaps add to uncertainty.