Friday, April 17, 2026
Daily picks
28
articles scored
#1 GOLDReleaseAnthropic News
Introducing Claude Opus 4.7
- Anthropic released Claude Opus 4.7 — a direct upgrade to Opus 4.6 at the same price, with stronger coding, better multi-step agentic workflows, and significantly improved vision
- SWE-bench Verified jumps from 80.8% → 87.6%; Cursor's internal coding benchmark goes from 58% → 70%; tool error rate in complex workflows drops by two-thirds
- Vision gets a 3x resolution upgrade — images up to 3.75MP (up from 1.15MP) — so screenshots, diagrams, and dense documents are processed at much higher fidelity automatically, no API changes needed
- Opus 4.7 is the most literally instruction-following Claude model yet — worth auditing prompts written for earlier models before migrating
- Available today via Anthropic API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry
#2 SILVERReleaseClaude Code Releases
v2.1.111
- /ultrareview is new: comprehensive cloud-based code review using parallel multi-agent analysis — run with no args for your current branch, or /ultrareview <PR#> to review a specific GitHub PR
- xhigh effort level is live for Opus 4.7 — sits between high and max, now tunable via an interactive /effort slider with arrow-key navigation
- Auto mode is on by default for Max subscribers on Opus 4.7 — the --enable-auto-mode flag is no longer needed
#3 BRONZEReleaseGitHub anthropics/claude-agent-sdk-python
[Release] anthropics/claude-agent-sdk-python: v0.1.60
- list_subagents() and get_subagent_messages() are new session helpers for inspecting subagent message chains spawned during a session — no more manual transcript parsing to debug multi-agent workflows
- OpenTelemetry distributed tracing added: install with pip install claude-agent-sdk[otel] to propagate W3C trace context from the SDK to the CLI subprocess, connecting spans end-to-end