Signing you in...

Please wait while we verify your authentication

MorningMail.ai
Sign inGet started
Article · Thursday, September 3, 2026

AI developer tools · What shipped

For a senior engineer who already reads HN. Real changes in AI developer tools today: releases with version numbers, papers with benchmarks, repos that crossed a threshold worth knowing. Skip hype threads, pre-announcement leaks, and recycled summaries. Always link primary sources.

By Marius BongartsTech66 editions
← See today's latest
Editions
20 / 66
TUE22SEPTodayMON21SEPSUN20SEPSAT19SEPFRI18SEPTHU17SEPWED16SEPTUE15SEPMON14SEPSUN13SEPSAT12SEPFRI11SEPTHU10SEPWED9SEPTUE8SEPMON7SEPSUN6SEPSAT5SEPFRI4SEPTHU3SEPWED2SEPTUE1SEPMON31AUGSUN30AUGSAT29AUGFRI28AUGTHU27AUGWED26AUGTUE25AUGMON24AUGSUN23AUGSAT22AUGFRI21AUGTHU20AUGWED19AUGTUE18AUGMON17AUGSUN16AUGSAT15AUGFRI14AUGTHU13AUGWED12AUGTUE11AUGMON10AUGSUN9AUGSAT8AUGFRI7AUGTHU6AUGWED5AUGWED29JULTUE28JULMON27JULSUN26JULSAT25JULTHU16JULWED15JULTUE14JULMON13JULSUN12JULSAT11JULFRI10JULTHU9JULWED8JULTUE7JULMON6JULSUN5JUL
Generated by AI overnight from public sources, refreshed daily.
AI developer tools · What shipped
Thursday, September 3, 2026
AI developer tools · What shipped

Gemini 3.8 Flash ships, Claude Code v2.1.259, dev tools consolidate

1 min read

Gemini 3.8 Flash

Google's fastest reasoning model just got smarter without the price tag.

Gemini 3.8 Flash matches 3.7 Flash speed and cost ($0.75/$3.75 per million tokens) while lifting reasoning benchmarks across software engineering, STEM, and multi-step tasks [Source: Google Blog]. On DeepSWE v1.1 (long-horizon code problems), 3.8 Flash outperforms most frontier models. HLE-Verified reasoning hits 54.9% across professional domains—a meaningful jump from earlier versions.

Agentic routing economics just shifted again.

Gemini 3.8 Flash Cyber

Google's fortress for vulnerability hunting just opened the gates.

Gemini 3.8 Flash Cyber, available through the new Fairwind Program for authorized security teams, delivers 70%+ success rate across 20 programming languages for bug discovery [Source: Google Blog]. It hits 47.2% pass@1 on CWE-Bench for automated patching while costing 2.3–5.2× less than competing frontier models. Chrome Security found it produces 2.6× more correct patches than best-in-class commercial tools.

Security workflows just gained a credible frontier alternative.

Claude Code v2.1.259

Claude Code's terminal agent gains organizational muscle and audit trails.

The latest release introduces managed MCP servers, letting orgs centralize HTTP/SSE endpoints for all users, plus --permission-prompts none for unattended headless deployments [Source: GitHub]. GitLab merge request recognition joins GitHub support. Under the hood: concurrent session corruption is patched, prompt cache now invalidates on OAuth refresh, and tool-call workflows stabilize.

Enterprise deployment just became less fragmented.

Sources
Introducing Gemini 3.8 Flash and 3.8 Flash Cyber - Google Blog
Introducing Gemini 3.8 Flash and 3.8 Flash Cyber - Google Blog
13 hours ago ... ... models in benchmarks like Vals Finance Agent V2 and Harvey's Legal Agent Benchmark. ... AI Studio that generates realistic Three.js renderings of ...
blog.google
AI Summary

Google released Gemini 3.8 Flash, their best reasoning and coding model to date, maintaining the same speed and low cost of 3.7 Flash at $0.75 per million input tokens and $3.75 per million output tokens. On DeepSWE v1.1 benchmark for long-horizon software engineering, 3.8 Flash outperforms most larger frontier models while costing significantly less, and achieves 54.9% on HLE-Verified for multi-step reasoning across STEM, humanities, and professional fields. Gemini 3.8 Flash Cyber, available through the new Fairwind Program for trusted defenders, demonstrates frontier-level cybersecurity performance with 70%+ success rate in vulnerability discovery across 20 programming languages and 47.2% pass@1 on CWE-Bench for automated patching, while being 2.3-5.2x cheaper than competing frontier models; Chrome Security found it produces 2.6x more correct patches than the best commercial models.

Visit source
Releases · anthropics/claude-code - GitHub
Releases · anthropics/claude-code - GitHub
4 hours ago ... ... tools ), keeps file tools inside the working directory, refuses ... Updated the /claude-api skill with Admin API coverage (organization members ...
github.com
AI Summary

``` Claude Code v2.1.259 released with managed MCP servers support, allowing organizations to provide HTTP/SSE MCP servers to all users. Added --permission-prompts none for unattended headless hosts and GitLab merge request recognition. Major fixes include concurrent session state corruption, bash permission checks for file operands, prompt cache invalidation on OAuth token refresh, and numerous tool call and workflow improvements. v2.1.257 added Claude Fable 5.1 model as default with 1M context at $10/$50 per million tokens, new time format settings, and Containment Escape rule for auto mode. v2.1.251 introduced PreModelSwitch/PostModelSwitch hook events, live streaming of foreground subagent tool calls to Remote Control clients, and per-session prompt-cache metrics in /cost. v2.1.248 added --restricted mode removing command/code execution tools, experimental.cacheTtl for agent-level prompt cache TTL, and cross-session messaging on Bedrock/Vertex/Foundry. v2.1.247 added SendFeedback tool for Claude to draft feedback reports, /claude-api cost-optimize command for profiling Claude API spend, and updated /claude-api skill with Admin API coverage for organization management. ```

Visit source
Compiled overnight by MorningMail.aiDelivered at 05:10 AM