New `Claude Opus 4.8` Is Trained to Be More 'Honest' and Stop 'Jumping to Conclusions'

// related

• #0001

  #0001Models & APIGoogle AI1 hour ago

  `Gemini Omni` and `Gemini 3.5` demos released after Google I/O 2026

  100radar

  Google is showing multimodal and model-upgrade capabilities through 11 videos. Treat this as an evaluation queue, not a spec sheet; API pricing and availability decide adoption.

  • —The demos cover Gemini Omni and Gemini 3.5, both announced at Google I/O 2026; this is the first hands-on signal after launch.
  • —Video demos are useful for UX pattern discovery: voice, vision, and real-time flows can map directly to app prototypes.
  • —No pricing, limits, latency, or API availability details are included here. Production decisions still need official docs.
  • —For solo products, the near-term value is competitive scanning: identify demos that can become narrow paid features.

  Source: blog.google/innovation-and-ai/models-and-research/geminiRead original →
• #0002

  #0002Models & APIGeekNewsyesterday

  Anthropic Ships `Claude Opus 4.8`

  90radar

  Same-price upgrade over 4.7, with better benchmarks and collaboration. Worth re-testing as a coding and agent default before changing spend assumptions.

  • —Built on Claude Opus 4.7; this is an incremental top-tier model refresh, not a new product line.
  • —Benchmark performance improved while pricing stays unchanged. Existing API cost models do not need recalculation first.
  • —Collaboration capability was strengthened, so agent workflows and pair-programming style tasks deserve fresh evals.

  Source: news.hada.io/topic?id=29960Read original →
• #0003

  #0003Models & APIAnthropic Python SDKyesterday

  `anthropic-sdk-python` `v0.105.0` adds `claude-opus-4-8` support

  70radar

  anthropic-sdk-pythonPython SDK — official client for Anthropic API

  Python clients can now call the new Opus variant and inspect richer token details. Worth upgrading before wiring new Claude features into production.

  • —claude-opus-4-8 is now supported in the official Python SDK, removing manual model-name workarounds.
  • —Mid-conversation system blocks landed in the API layer, useful for agent flows that need runtime instruction changes.
  • —usage.output_tokens_details exposes finer output-token accounting, making cost/debug dashboards easier to build.
  • —Custom file size caps are supported, so upload limits can be enforced closer to the SDK boundary.

  Source: github.com/anthropics/anthropic-sdk-python/releases/tag/Read original →