The GitHub Blog · Oct 6, 2026 · 5:21 AM

GitHub opens ReviewBench, a public benchmark you can run your own AI code reviewer against

The research-preview benchmark scores code review agents on 219 real public PRs across 19 languages, with the full dataset, rubric, and Claude Sonnet 5 judge published, and you can sign in with GitHub, bring a container image plus your own model key, and test on a 25-PR set before a full leaderboard run. GitHub built it to tune Copilot code review, the golden set leans on LLM judging, and leaderboard entries need maintainer approval, so treat the rankings as a starting point.

Read at github.blog

GitHub Copilot

AI coding assistance across editors, GitHub, and the CLI

More news

View all

Copilot CLI 1.0.92 adds `copilot config` commands and stops passing GITHUB_TOKEN into sandboxed shells

The Oct 5 release adds copilot config subcommands to list, read, set, and remove settings, a Ctrl+E picker before a conversation to choose a local or cloud run, and live stdout/stderr streaming for shell tool calls. Sandboxed shells no longer inherit your ambient GITHUB TOKEN unless you configure it, so scripts that quietly relied on it will need it set explicitly…

GitHub / copilot-cli releases 7 minutes ago

Superagent open-sources Security-One 27B, a model that scores agent tool calls and inputs as safe or unsafe

Instead of writing out an answer, the Apache 2.0 model returns probabilities for answers you define (like safe and unsafe ) for a prompt, tool call, code change, or alert, so your own code picks a threshold to allow, block, or escalate; Superagent reports catching 599 of 600 BIPIA prompt injections. Prompt injection screening is the best tested use so far, it flagged 12% of tricky benign NotInject inputs, and the numbers are Superagent's own, so tune it on your own traffic and keep real permission checks outside the model…

Superagent Blog 2 hours ago

Codex Day 1: GPT-6 Astra and GPT-6.1 Sol run about 50% faster on subscriptions, partners included

The first drop of OpenAI's 28 day improvement or reset run is an inference speedup of roughly 50% for GPT 6 Astra and GPT 6.1 Sol when used through a ChatGPT subscription, including partner tools that use Sign in with ChatGPT like OpenCode, Pi, Amp, and Devin, with nothing to change on your end. API speeds aren't part of it, and one early forum benchmark saw 6.1 Sol get slower mid rollout, so check your own runs…

OpenAI Developer Community 3 hours ago

GitHub secret scanning now catches leaked Lovable API keys and Supabase access tokens

As of Oct 5, secret scanning flags lovable api key , Supabase OAuth and scoped personal access tokens, and Pydantic Logfire and AI Gateway keys in your repos, and Lovable joined the partner program, so a Lovable key pushed to a public repo gets sent to Lovable to revoke or rotate. If your agent ever committed a .env , check your repo's security alerts…

GitHub Changelog 4 hours ago

Reflection previews Beam, a 501B open-weight MoE coding model with Apache 2.0 weights due this month

Beam activates 23B of its 501B parameters per token, has 1M token context, and Reflection reports 80.9 on SWE bench Verified and 80.1 on Terminal Bench 2.1, roughly level with GLM 5.2 at 3–4x less inference compute but behind Kimi K3 and GLM 5.3. It's waitlist only for now, with weights, a technical report, and fine tuning tooling promised later in October…

Reflection Blog 5 hours ago

Spotted something we missed? Start a thread.