{
  "id": 9260934,
  "title": "Kinetic-4B vs Claude Haiku 4.5: The 4B Model Wins Tools",
  "url": "https://urgent.news/2026/09/23/kinetic-4b-vs-claude-haiku-4-5-the-4b-model-wins-tools",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-23T03:43:49.000Z",
  "source": {
    "name": "Dev.to",
    "slug": "dev-to",
    "url": "https://dev.to/shaam_ai/kinetic-4b-vs-claude-haiku-45-the-4b-model-wins-tools-53h4"
  },
  "original_language": "en",
  "account": "The Kinetic-4B model, developed by Bengaluru-based Conscious Engines, outperformed Anthropic's Claude Haiku 4.5 in a recent tool calling benchmark. Kinetic-4B achieved 82.33% accuracy, 95.33% tool-name accuracy, 4.67% failed calls, and a p95 latency of 1.61 seconds, compared to Claude Haiku 4.5's 80.0% accuracy, 90.33% tool-name accuracy, 9.67% failed calls, and a p95 latency of 4.02 seconds. In contrast, OpenAI's GPT-OSS-120B trailed with 76.33% accuracy and a p95 latency of 7.99 seconds. The smaller Kinetic-4B model, trained on a single rented GPU for about 4.5 hours, demonstrates that fine-tuned, smaller models are more cost-effective and faster for specific tasks like tool calling, while larger models excel at open-ended reasoning and code generation.",
  "summary": "Kinetic-4B wins on tool calling. On a 300-sample Composio evaluation, the 4-billion-parameter model from Bengaluru lab Conscious Engines scored 82.33% accuracy at 1.61s p95 latency, against 80.0% and 4.02s for Anthropic's Claude Haiku 4.5, and 76.33% at 7.99s for OpenAI's GPT-OSS-120B ( Conscious Engines, 1 April 2026 ). That is 2.5x lower tail latency at slightly higher accuracy. Haiku remains…",
  "key_points": [
    "Kinetic-4B model outperforms Claude Haiku 4.5 in tool calling benchmark",
    "Achieves 82.33% accuracy, 95.33% tool-name accuracy, 1.61-second latency",
    "Smaller models cost-effective and faster for specific tasks like tool calling"
  ],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}