flashtea/
paper

No. 7 · Friday, 19 June 2026 · 7 stories

AI for Software Development

Grok 4.3 reaches Amazon Bedrock with 1M-token context and configurable reasoning

xAI's Grok 4.3 reached general availability on Amazon Bedrock on June 16, giving AWS developers access to a 1-million-token context window and three reasoning effort levels (low, medium, high) via the standard Bedrock API. On-demand pricing is $1.25 per million input tokens and $2.50 per million output tokens; inference runs on Amazon's new Mantle engine with a 30,000-token maximum output per call.

why it mattersAWS developers can now benchmark Grok 4.3 against Claude and other Bedrock models in a single API call — at pricing that undercuts the direct xAI API for most workloads, with no separate account required.

AWS

AI for Software Development2 stories, lead above

Claude Code v2.1.181: /config prompt command, Bun 1.4 runtime, and mid-thinking auto-retry

The June 17 Claude Code release adds /config key=value syntax to set any configuration value directly from the prompt — works in interactive, -p, and Remote Control modes. The bundled Bun runtime was upgraded to 1.4, long-paragraph streaming was improved to appear line-by-line, and API connections that drop mid-thinking now auto-retry instead of surfacing a raw error. Fixes land for JetBrains IDE flickering on 2026.1+ and Shift+non-ASCII key drops in Kitty-protocol terminals (WezTerm, Ghostty, kitty).

why it matters/config from inside the session eliminates the need to exit to settings for quick per-session overrides like toggling thinking mode or switching the active model.

Anthropic (Claude Code changelog)

AI Research & Models1 story

Grok 4.3 model card: configurable reasoning depth and Grok 5 at 10T parameters confirmed on roadmap

Alongside its Bedrock debut, xAI published Grok 4.3's full model card: 1M-token context, 30K max output, and reasoning configurable at low, medium, or high effort — each level progressively more compute-intensive. The model is positioned for enterprise workloads including contract review, case-law research, and financial document Q&A. xAI separately confirmed Grok 5 is in training at an estimated 10 trillion parameters, with no public release timeline.

why it mattersPer-call reasoning-effort control is now standard across Claude, GPT-5.5, and Grok — making cost-per-task optimization a routine engineering decision rather than a model-selection one.

xAI

Dev Tools & Releases2 stories

npm v12 will block install scripts, Git dependencies, and remote URLs by default — July cutover deadline

GitHub's June 9 changelog confirmed npm v12 will flip three historically permissive defaults: blocking preinstall/install/postinstall scripts from dependencies, Git-sourced packages, and remote-URL package sources. The change was accelerated by the Miasma supply chain attack of June 1, which compromised 89 packages across Red Hat's npm namespace in two hours. All three restrictions surface as warnings today in npm 11.16.0; v12 itself is due in July 2026.

why it mattersAny CI pipeline relying on install scripts or Git-sourced dependencies will break silently on upgrade day unless teams run npm 11.16.0 now and clear the warning output.

GitHub Changelog

VS Code 1.124 enables Copilot Autopilot by default with a three-loop safety cap

VS Code 1.124, released June 10, turns on Copilot Autopilot for all users by default — the agent now writes files, executes terminal commands, and responds to tool prompts without per-step confirmation. A new Advanced Autopilot mode uses a small utility model to determine when the task is genuinely complete rather than relying on fixed rules; autonomous loops are capped at three before Autopilot halts. Organizations can restrict the behavior via the chat.tools.global.autoApprove setting.

why it mattersAutopilot on by default is a behavioral shift in the world's most popular editor; the three-loop cap and org-level policy gate offer meaningful guardrails for teams concerned about runaway agents.

Visual Studio Magazine

Bitcoin2 stories

BTC stabilizes near $64k; Fear & Greed at 22 (extreme fear) as CME prices 61% odds of October rate hike

Bitcoin has settled in the $63,500–$65,000 range in the 48 hours since the Federal Reserve's June 17 FOMC decision held rates at 3.50–3.75% but issued its most hawkish dot plot in two years — nine of eighteen officials projecting at least one hike before year-end. The Crypto Fear & Greed Index fell to 22, the lowest reading since February. CME FedWatch now prices a 61% chance of an October hike, effectively removing the dovish macro backstop that underpinned BTC's 2025 rally.

why it mattersChair Warsh's deliberate refusal to submit a dot projection and stripped forward-guidance language signal a genuinely unpredictable rate path — a structural headwind for all risk assets through at least Q3.

CCN

Fidelity's FBTC takes $14M in inflows on June 17 as sector-wide Bitcoin ETF outflows total $111M

Despite Bitcoin spot ETFs collectively shedding $111 million in net outflows on June 17, Fidelity's FBTC was the only major issuer to post positive flows on the day, attracting $14 million in fresh institutional inflows. Cumulative outflows over the past three weeks have exceeded $3.5 billion — the largest sustained withdrawal streak since the products launched in January 2024.

why it mattersSelective accumulation at Fidelity while peers bleed suggests differentiated institutional conviction rather than a uniform exit — a pattern that has historically preceded stabilization.

CoinDesk

Bitcoin · weekly close vs 200-week moving average · log scale · last 8 years Price200-week average
$5,000$10,000$20,000$50,000$100,000 20192020202120222023202420252026 $80,809 $67,358
$5,000$20,000 2020202220242026 $80,809 $67,358

Price is 1.20× the 200-week average. Weekly closes, USD, as of 28 October 2026 · data blockchain.com.