The Builder's Radar
Connecting builders & tracking AI frontier
USA vs China
dev metrics · updated 13 Oct 2026 · click a row to fly ·
Global inference share · text requests
US models · OpenAI · Google · Anthropic
CN models · DeepSeek · Z.ai · Qwen · Xiaomi · Tencent
China still leads usage but the gap narrowed to 7.6 points (47.9% vs 40.3% of OpenRouter text requests, w/e 5-oct-2026) — and the volume crown is no longer 100% China: DeepSeek V4.1 Flash tops the world at 113T tokens/month, with US models GPT-5.6 Luna and Nemotron inside the top 10.→ CHINA
Frontier capability · Vals Index
best US model · Gemini 4 Argon
best CN model · MiMo V2.6 Pro
Two fronts, two winners: China still leads usage (47.9% vs 40.3% of OpenRouter requests) but the US takes capability — 13.7 points between the best model of each region on Vals' independent leaderboard. Snapshot 13-oct-2026 (v2.1): they re-run the index as models are added, so the values move between visits.→ USA
Frontier training run · public cost
US runs · Grok 4 ≈ $388M (Epoch)
MiMo-V2.6 · RL phase (live log)
Xiaomi's entire livestreamed RL bill (Pro + Flash) is ~1/30 of one US frontier run — scopes differ, the efficiency gap doesn't.→ CHINA
Flagship API cost · $/1M input tokens
GPT-6 Astra
DeepSeek V4 Pro
7.6x cheaper at the flagship tier — China keeps the floor under frontier pricing.→ CHINA
Value-tier API cost · $/1M input tokens
Gemini 3.8 Flash
MiMo-V2.6-Flash (Xiaomi)
5.4x under Google's fast tier — Xiaomi's 309B/15B MoE (MIT, 1M ctx) prices for volume, not margin.→ CHINA
Open-weight flagships · tracked
GPT-6 · Claude · Gemini closed
DeepSeek · Qwen · GLM open
Every tracked Chinese flagship ships open weights (MIT · Qwen & Z.ai lab licenses); all three US flagships stay closed.→ CHINA
Max flagship context window
GPT-6 Astra
MiMo-V2.6 (Xiaomi)
Dead tie at 1.05M: Astra got there first (3-sep) and MiMo-V2.6 matched it (22-sep). RAG builders: both fit a mid-size codebase.→ USA
Consumer AI adoption · national stats
adults 18–64 · US RPS survey (May 2026)
share of population · CNNIC (H1 2026)
Denominators shown on purpose (they differ by design): the US surveys working-age adults, China counts the whole country — 700M+ users and past 50% penetration in H1-2026. US leads intensity, China leads velocity.→ USA
Latest LLM Releases
Muse Wonder
Meta's frontier reasoning model for long agentic tasks — omnimodal in (text, image, video, audio, PDF), text out, 1M context.
Gemini 4 Argon
Google's new flagship reasoning model: built for complex software engineering, math/science and autonomous multi-step agents — 1M context.
GPT-6.1 Sol · Pro
OpenAI steps its Sol line to 6.1 with a Pro tier alongside — closed weights, API only.
Claude Sonnet 5.5
Anthropic's mid tier moves to 5.5 a week after Opus 5.5: 1M-token context, 128k max output, $2/$10 per 1M tokens.
MiMo-V2.6 Pro · Flash
Xiaomi's omnimodal MoE pair ships open (MIT): Flash is 309B/15B active at $0.14/M input with 1M context.
Claude Opus 5.5
Anthropic's flagship for demanding reasoning, coding and long-horizon agentic work — succeeding Opus 5, strong on multi-step changes in large codebases.
DeepSeek V4.1 Flash
Newest Flash-tier open release — DeepSeek keeps compressing frontier quality into cheap, fast inference.
GPT-6 Astra
OpenAI's new flagship line, shipping alongside Astra Pro on day one.
Qwen3.8 Max (0902)
Alibaba refreshes its multimodal APAC flagship with the 0902 build.
Gemini 3.8 Flash
Google's fast multimodal tier steps to 3.8 — best price/perf in the Flash line.
Claude Fable 5.1 · Mythos 5.1
Anthropic's frontier agentic coding line gets a point release — Fable for work, Mythos at the ceiling.
GLM-5.3 Flash
Zhipu's self-hostable frontier flash — the fastest open challenger you can run yourself.
Coding agents
○ curatedOpen-source coding agent that runs in your own terminal.
anomalyco · anomalyco/opencode
Anthropic's agentic coding assistant, straight from the shell.
Anthropic · anthropics/claude-code
OpenAI's coding agent — open source, written in Rust.
OpenAI · openai/codex
Google's terminal agent with a free allowance for individuals.
Google · google-gemini/gemini-cli
Open agent that edits, runs and tests — not only suggests.
OpenHands (ex All-Hands AI) · OpenHands/OpenHands
Autonomous agent that lives inside VS Code.
Cline · cline/cline
Extensible agent that installs, executes and tests on your machine.
Block · aaif-goose/goose
Pair programming in the terminal, committing straight into git.
Aider AI · Aider-AI/aider
Collaborative software-agent teams (CN), open research stack.
OpenBMB · Tsinghua · OpenBMB/ChatDev
Multi-agent framework from Alibaba, built for real deployments.
Alibaba · agentscope-ai/agentscope
★ straight from the GitHub API · ○ = curated · if GitHub doesn't answer, the row shows ★ — (nothing estimated here)
GitHub Trending
liveLoading...
Hacker News
liveLoading...
Radar Stats
Sample data. Real builders replace samples as they join.
One builder runs Unlinked. Free for builders, forever — funded by sustainers and tool grants, never by ads.
Fuel the Radar →