Claude Opus 4.7 Model Card Released
Anthropic has released the Claude Opus 4.7 model card, which serves as technical documentation for their AI model. The document appears to be a PDF containing system specifications, though the provided content shows binary/encoded data rather than readable text.
From the source context, we can see this is a model card release that has generated discussion on Hacker News with 118 points and 55 comments. Model cards typically include technical specifications, capabilities, limitations, and safety considerations for AI models.
For developers working with AI coding agents, model cards provide essential information about:
- Model capabilities and limitations
- Performance characteristics
- Safety mitigations
- Intended use cases
- Technical specifications
Since the provided PDF content is encoded, readers should refer to the actual source document for specific technical details about Claude Opus 4.7's features, benchmarks, and system requirements.
📖 Read the full source: HN AI Agents
👀 See Also
Anthropic Doubles Claude Code Usage Limits, Signs SpaceX Compute Deal
Anthropic doubled five-hour usage windows for Claude Code Pro and Max subscribers, removed peak-hour reductions, and raised API limits for Opus, citing a new deal with SpaceX for 300+ MW of compute capacity from the Colossus 1 supercomputer (220,000+ NVIDIA GPUs).
Claude Code v2.1.132: SIGINT Graceful Shutdown, MCP Fixes, and Terminal Handling Overhaul
Claude Code v2.1.132 fixes graceful shutdown on external SIGINT, adds CLAUDE_CODE_SESSION_ID and CLAUDE_CODE_DISABLE_ALTERNATE_SCREEN env vars, patches MCP memory leaks and tool listing retries, and resolves dozens of terminal edge cases across IDE terminals.
Qwen3.6-27B Fits on Single 24GB GPU, Beats Former 397B MoE on SWE-bench
Qwen3.6-27B (Apache 2.0, 262K context) runs at Q4_K_M in ~16.8GB, achieving SWE-bench Verified 77.2 — outperforming Qwen3.5-397B-A17B MoE (76.2). Uses Gated DeltaNet linear attention with Thinking Preservation for agent workflows.
Talkie: A 13B LLM Trained Exclusively on Pre-1931 Text, Using Claude as a Judge in RL Training
Researchers released Talkie, a 13B LLM trained only on text published before 1931 (no internet, no WWII data). Claude Sonnet 4.6 was used as the judge in its online DPO reinforcement learning pipeline, and Claude Opus 4.4 generated synthetic multi-turn conversations for fine-tuning. The model can write Python code from a few in-context examples despite zero modern code in training.