Writing about claude-multi.
Build notes, deep dives, and the occasional rant. New posts whenever there's something worth saying.
- · by hmziqrs · #providers#models#glm#benchmark#claude-code
GLM-5.2 for Claude Code: 1M Context and Three-Tier Model Mapping
GLM-5.2 brings a 1,000,000-token context window to the Z.ai Coding Plan. The claude-multi GLM template now maps it to the opus slot, with GLM-5.1 as sonnet and GLM-5-Turbo as haiku.
- · by hmziqrs · #providers#models#kimi#benchmark#claude-code
Kimi K2.7 Code in Claude Code: Setup, Benchmarks & Tiers
Kimi K2.7 Code is now available in Claude Code via claude-multi. See benchmark results, the new three-tier model mapping (opus/sonnet/haiku), and setup steps.
- · by hmziqrs · #release#bugfix#build
v0.8.1: Fixing the Broken npm Package
If you installed claude-multi globally and got Module not found, this is why. The CLI build and the docs site were both outputting to dist/. The docs won. Now they don't.
- · by hmziqrs · #release#sync#plugins#web#responsive
v0.8.2: Granular Sync Modes and Responsive Web
claude-multi had two sync modes: symlink everything or copy everything. Now there's a middle ground. Plus a responsive header, sticky sidebar, and SSR footer for the docs site.
- · by hmziqrs · #providers#models#minimax#benchmark#claude-code
MiniMax M3 for Claude Code: 1M Context, Benchmarks, Pricing
MiniMax M3 brings a 1M-token context window, frontier coding scores, and native multimodality to Claude Code. Updated template, benchmarks vs Opus 4.7, and pricing.
- · by hmziqrs · #release#health#migration#ui#hardening
v0.6.5: Action Buttons, Health Screen Fix, Hardened Migrations
Instance details now has action buttons to update settings templates and regenerate alias wrappers. The health screen got a rewrite to fix broken dismiss actions. Migrations are more resilient with stale lock detection and atomic health status writes.
- · by hmziqrs · #release#architecture#health#migration#bugfix
v0.7.0: No More Pinned Claude Binary
claude-multi no longer bundles its own copy of Claude Code. All instances now use your globally installed claude binary, which means they auto-update together. Doctor fix, health checks, and migrations were rewritten. Four pre-existing bugs were fixed along the way.
- · by hmziqrs · #Claude Code#third-party providers#compatibility#claude-multi
Claude Code v2.1.156 fixes the third-party provider breakage
Anthropic shipped v2.1.156, which fixes the HTTP 422 error that broke every non-Anthropic API provider in v2.1.154 and v2.1.155. Here is what changed and what claude-multi users need to do.
- · by hmziqrs · #release#migration#providers#templates
v0.6.3: Drop the Pinned Binary, Update Every Provider Template
Instance migrations now point wrappers at your global claude install instead of a stale pinned binary. Every provider template got correct thinking and output token limits.
- · by hmziqrs · #release#migration#providers#templates
v0.6.4: Existing Instances Now Auto-Sync Provider Template Updates
A new instance migration detects which provider each instance uses and re-applies the latest template config (model names, thinking tokens, output limits) to settings.json. API keys and custom env vars are preserved.
- · by hmziqrs · #Claude Code#post-mortem#third-party providers#compatibility#claude-multi
Claude Code v2.1.154 Broke Every Third-Party Provider. Here's What Happened.
On May 28, 2026, Anthropic shipped a Claude Code update that broke every non-Anthropic API provider within hours. A post-mortem on what went wrong, the fallout, and the compatibility safeguards we added to claude-multi.
- · by hmziqrs · #Claude Code#AI engineering#LLM integration#claude-multi#developer tools#AI coding#workflow automation#MCP
Claude Code is doing more of the job now, and claude-multi makes the rest of it cheaper
Claude Code has moved past autocomplete. With MCP and a million-token context, it handles real workflows end to end. claude-multi is how you point it at different providers without burning your config to the ground.
- · by hmziqrs · #Claude Code#Model Context Protocol#MCP#workflow automation#AI co-engineer#developer productivity#tool integration#AI in software development
How MCP lets Claude Code actually do the rest of your job
MCP gives Claude Code a way to talk to the tools you already use: Jira, GitHub, Slack, your databases. Here is what that buys you and where it breaks down.
- · by hmziqrs · #providers#templates#announcement
Five New Provider Templates: MiMo, Kimi, Qwen, and More
claude-multi now ships templates for Xiaomi MiMo, Moonshot Kimi, and Alibaba Qwen, including separate templates for providers that split their API across pay-per-token and subscription coding plans.
- · by hmziqrs · #LLM cost optimization#AI development#LLM routing#claude-code-llm-router#tiered models#cost savings#AI engineering#developer tools
Stop paying Opus prices for a git status
If you're sending every request to a flagship model, you're overpaying by a lot. A look at how LLM routing actually works, what kind of savings to expect, and how to set it up.
- · by hmziqrs · #AI coding#LLM comparison#Xiaomi MiMo#Moonshot Kimi#Alibaba Qwen#agentic AI#open-source LLMs#developer tools#claude-multi#workflow automation
Three coding models worth paying attention to: MiMo, Kimi, and Qwen
Xiaomi MiMo-V2.5-Pro, Moonshot Kimi K2.6, and Alibaba Qwen3-Coder-Next are doing real work now. A look at what each is actually good at, and how to wire them up with claude-multi.
- · by hmziqrs · #claude-code#tui#deep-dive
Inside claude-multi: a tour of every menu
A walkthrough of every screen in the claude-multi TUI. What each option does, when to use it, and the design calls I made along the way.
Building software for 9 years. TypeScript, Rust, Flutter. Writing about things I broke and fixed.