AI Weekly Malaysia

OpenRouter has cut GPT-5.6 Sol pricing by 50%, dropping input to $2.50/1M tokens and output to $15/1M tokens (from $5.00/$30.00), with cache reads halved to $0.25/1M. The model has a 1M token context window, was released Jul 9 2026, and is served by OpenAI, Azure, and Amazon Bedrock with throughput up to 61 tok/s on Bedrock US and 99.22%+ uptime across providers.

Why: If you're running GPT-5.6 Sol in production via OpenRouter, your API costs just dropped by half — re-evaluate your routing mode (Balanced vs Nitro vs Exacto) since Bedrock US offers 61 tok/s throughput at a slightly higher $5.50/$33 list price but may still be cheaper post-discount for latency-sensitive workloads. The $0.25 cache read price makes prompt caching materially more valuable for agentic and multi-step coding workflows.

Roboflow benchmarked OpenAI's GPT-5.6 lineup (Sol, Terra, Luna) on vision tasks including detection, counting, OCR, and data extraction. Sol scored 46.2 mAP@50 on object detection versus GPT-5.5's 13.8, turning detection from a weakness into a practical capability, though the model sometimes hallucinated boxes in unnatural grid layouts.

Why: If you're building vision pipelines with GPT-5.6, prompt for absolute XYXY coordinates in image pixels—using the wrong format (e.g., Gemini-style normalized YXYX) costs ~15 mAP points. Sol is now viable for document layout detection and dense-scene counting, but expect occasional phantom boxes with no ground-truth overlap, so you still need post-processing validation.