🚨 Breaking
None
🗑️ Dépréciations
None
💰 Pricing
No pricing changes reported today.
🆕 Nouveautés
- Anthropic launched Claude Opus 4.8 (
claude-opus-4-8) with 1M token context, 128k max output, same tooling as Opus 4.7. - vLLM v0.23.0 released with DeepSeek-V4 maturity across backends, 408 commits from 200 contributors. Release notes
- SGLang v0.5.13 adds Nemotron 3 Ultra day-0 support, Step-3.7-Flash, Command A+. Release
- llama.cpp multiple builds (b9611–b9628) focusing on CI fixes, UI asset cleanup, and SYCL integration. Details
- OpenRouter introduces Fusion – multi-model deliberation with parallel web search and fetch.
- MoonshotAI releases Kimi K2.7 Code, a coding-focused MoE model for long-context tasks.
🌐 Actualité IA
No significant industry signals today.
💡 Conseil du jour
Evaluate vLLM v0.23.0 if you serve DeepSeek-V4; the maturation across backends may simplify your deployment. For coding workflows, test Kimi K2.7 Code via OpenRouter against your existing code models to gauge cost/quality trade-offs.