Anthropic published Claude Sonnet 5’s price as $3/$15 per million tokens — identical to Sonnet 4.6. But Sonnet 5 uses a new tokenizer that inflates token counts by 42% for English and 27% for Python, and generates roughly 300M output tokens per task. The real cost per task is $2.29 vs Sonnet 4.6’s $1.20. Here is the math developers need.
Fable 5 went offline on June 12 after a single Amazon engineer used it to push unauthorized infrastructure code. It came back July 1 with a 50% usage cap, a new safety classifier, and terms that let governments suspend frontier AI models in hours. Here is what the 19 days tell developers about the future of AI infrastructure.
Claude Sonnet 5 launched June 30 at 40% less per token than Opus 4.8. The surprising finding: at standard pricing, Sonnet 5 costs 15% more per task. Full benchmark breakdown on SWE-bench Pro, Terminal-Bench, speed, and the 12-scenario decision guide.
Claude Tag launched June 23 as a public beta for Enterprise and Team customers. It is not a chatbot — it is a persistent agent with ambient mode that already approves 65% of Anthropic’s own product team’s code changes. Here is what engineering teams need to know before August 3.
Z.ai, the creator of GLM models, recently launched GLM 5.2. It’s an open-weight model that most developers hadn’t followed this
Claude Mythos leads GPT-5.5 by 19 points on SWE-bench Pro and 25 points on BenchLM Coding — but it is not publicly accessible and costs 4x more per output token. Here is the full breakdown developers need before Mythos goes live.
Friday AI turns your PC into an autonomous product team. Combines the best models from Claude, OpenAI, Gemini, and ElevenLabs into one seamless operator that doesn’t just think, it executes.