Three independent developer-adoption signals, Jan 2025 – Jul 2026. Values are shown as
average per day within each month; toggle Share % vs Per-day volume. Jul 2026 is
partial (marked *, projected ×31/n for month totals).
Each signal is a biased proxy — read the caveat notes under each chart. Not a measure of production usage.
Key takeaways
Closed owns developer breadth; open owns volume. In actual code ~75% of repos call closed APIs (OpenAI/Anthropic/Google) vs ~25% open, and the anthropic SDK grew ~24×/day — even as open models passed ~69% of OpenRouter token volume. Breadth and volume point opposite ways.
Agentic coding exploded and is ~100% closed-tooled. AI agents committing as their own account went from ~0 to thousands/day through 2025 — entirely closed products (Copilot, Codex, Claude Code, Gemini). Open-model agents are near-invisible.
Open adoption is systematically undercounted here. Open models are called via the OpenAI-compatible client (not their own SDK), and open agents run under a human's account — so every signal below treats open as a floor, not a measure.
Part 1
Current snapshot — GitHub Code Search 2026-07-18
Public code files that reference each provider's API endpoint in live code — repos actually calling each model. A single live query on 2026-07-18 (a point-in-time count, not a monthly total or a per-day rate). The one breadth signal from real code.
Code files referencing each provider's API endpoint —
Caveats — read before interpreting.
Counts code files, not distinct repos (one repo can contain several); total_count is GitHub's estimate. Magnitude/ratio is solid, exact counts less so.
Endpoints only (e.g. api.deepseek.com) — clean & symmetric across camps. Llama excluded (no single API endpoint; it shows ~38.5k via a model-path query, not comparable).
Matches the string anywhere in code (incl. comments/vendored copies). Snapshot only — no history (that's why the trends above use PyPI / GitHub Archive instead).
Part 2
Adoption over time
How developer adoption evolved, Jan 2025 – Jul 2026 — SDK installs, GitHub payload mentions, and GitHub collaborators. Three time-series proxies; none measures production usage, and each is biased (read the caveat notes under every chart). Values are average-per-day within each month.
2.1 · PyPI SDK install
Developers installing each provider's Python SDK — an install-activity proxy.
openai is a contaminated ceiling — it's the client for OpenAI-compatible calls to DeepSeek/GLM/Qwen/local, not just OpenAI.
Open is undercounted: open models are called through the openai client, not their own SDK (zhipuai stayed flat while GLM usage rose).
"Open self-host infra" = transformers/vllm/sglang — broad, not model-specific. → open-vs-closed share here is directional only.
2.2 · GitHub payload mention
Distinct repos whose GitHub event payload text (issue / PR / commit message / comment) names a model — scanned from GitHub Archive event payloads, bots excluded. A discussion / mindshare proxy, not code.
Open vs closed
By vendor
Caveats — read before interpreting.
What "mention" means: a regex match of a model's API endpoint or a versioned model-id (e.g. api.deepseek.com, glm-4) in the event's payload text. It counts the string being present (in an issue, PR, commit message, or comment), not that the repo actually runs the model.
Why prose, not code: GitHub Archive has no code-file data (only event text); and the github_repos code dataset is frozen — contents last modified 2022-11 (only commits refreshed 2025-10). So 2025–26 code can't be grepped; we detect mentions in event payloads instead.
Prose mentions ~100× undercount real code usage (Code Search finds api.anthropic.com in ~45k code files vs a few hundred prose mentions/mo).
Bots excluded, but model-id regex noise remains → don't trust absolute cross-vendor rankings (Qwen > Anthropic here is an artifact).
Known artifact: the openai spike Jun–Oct 2025 is a mass-PR pattern (PR events 28k→682k), not adoption — ignore that window's levels. Use for within-vendor trends, not levels.
2.3 · GitHub collaborator closed-only
"Collaborator" = an AI coding agent committing / opening PRs as its own GitHub account (a collaborator on the repo): Copilot (GitHub, multi-model), chatgpt-codex-connector[bot] (OpenAI Codex), claude[bot] (Anthropic), gemini-code-assist[bot] (Google). Measures agentic-coding adoption — AI writing commits directly, not humans discussing models.
By vendor
Caveats — read before interpreting.
No open/closed split shown — the open camp is a flat 0. Every vendor-native coding agent is a closed product, so an open-vs-closed chart would just be one line. It is ~100% closed by construction, not because open models aren't used: open-model agents (aider/OpenHands) run locally and commit under the developer's own account, invisible here (would need commit author-email, which GitHub Archive lacks).
GitHub Copilot is its own line, not folded into OpenAI — it's a Microsoft/GitHub product that runs multiple models (GPT, Claude, Gemini), so it's not attributed to any single model vendor. "OpenAI (Codex)" is the actual OpenAI agent, and it's ~10× smaller than Copilot (Copilot 11.6M vs Codex 1.0M events over the window). Other products (cursor/coderabbit/devin) are excluded.
2026 tail-off is unverified (possible bot-login changes) — treat the most recent months cautiously.