作り始める前の無料 LLM API 検証チェックリスト
プロバイダーに依存する前に、無料 LLM API の利用条件・テストエビデンス・運用上の制限を確認するための開発者チェックリスト。
日本語で読める
記事を読む →ブログ · ガイド · 検証済みの AI ニュース
構築できることを変える AI の変化を追います。
申請手順・設定・安全性・無料枠の変化に役立つ一次情報を集めます。各記事の原文言語は明記します。
英語原文
一部の記事は日本語で読めます。それ以外は英語原文へのリンクです。
未翻訳の原文を、日本語の記事として見せることはありません。
OpenAI 互換 API：「そのまま置き換え可能」と呼ぶ前にテストすべきこと
リクエスト形状・レスポンス意味論・エラー・制限と、アプリケーションが実際に依存する機能のための実践的な互換性テスト。
日本語で読める
記事を読む →モデル数ではなくエビデンスで無料 AI API を比較する方法
検証されたアクセス、使える制限、互換性、失敗時の挙動、変更リスクで無料 AI API を比較します。モデルディレクトリの規模ではありません。
日本語で読める
記事を読む →モデルのリリースは移行計画ではない：切り替える前にテストすること
公開・プレビュー・エイリアス変更・終了通知を分類し、再現可能な契約テストで AI アプリケーションを移行します。
日本語で読める
記事を読む →GPT-5.6 と Gemini 3.7 Flash：日付を明記した API 移行チェックリスト
2026 年 8 月の発表を記録し、最新の公式資料を踏まえて更新。移行前にモデル ID、インターフェース、アカウント制限、実務テストを確認します。
日本語で読める
記事を読む →Ox Alpha のプレビュー後：正体・料金・データ保持を再確認
OpenRouter は Ox Alpha を Z.AI GLM-5.3-Flash として公開しました。利用前にモデルの経路、料金、データ保持条件を確認してください。
日本語で読める
記事を読む →無料 API の障害に耐える AI Agent の設計
Agent の状態、モデルルーティング、ツール権限、API キーを分離し、プロバイダー障害やクォータ変更で全体が止まらないようにします。
日本語で読める
記事を読む →画像・音声・動画アプリに使う API の選び方
テキストモデルの順位ではなく、入出力、転送方式、フォールバック要件からマルチモーダル API を選びます。
日本語で読める
記事を読む →XingChen: Xing4.0 29B A4B (free) is now listed as a free model on OpenRouter
XingChen: Xing4.0 29B A4B (free) (xingchen-agi/xing4.0-29b-a4b:free) was detected with zero prompt, completion and request pricing in the official OpenRouter catalog. Review the access paths before building on it.
Official sourceIndependent audit
英語原文を読む →NVIDIA: Nemotron 3.5 Super VL (free) is now listed as a free model on OpenRouter
NVIDIA: Nemotron 3.5 Super VL (free) (nvidia/nemotron-3.5-super-vl:free) was detected with zero prompt, completion and request pricing in the official OpenRouter catalog. Review the access paths before building on it.
Official sourceIndependent audit
英語原文を読む →inclusionAI: Ling 3.1 Flash is now listed as a free model on OpenRouter
inclusionAI: Ling 3.1 Flash (inclusionai/ling-3.1-flash) was detected with zero prompt, completion and request pricing in the official OpenRouter catalog. Review the access paths before building on it.
Official sourceIndependent audit
英語原文を読む →Apodex: Apodex 1.1 Mini (free) is now listed as a free model on OpenRouter
Apodex: Apodex 1.1 Mini (free) (apodex/apodex-1.1-mini:free) was detected with zero prompt, completion and request pricing in the official OpenRouter catalog. Review the access paths before building on it.
Official sourceIndependent audit
英語原文を読む →Space Bunny Alpha is now listed as a free model on OpenRouter
Space Bunny Alpha (stealth/space-bunny-alpha) was detected with zero prompt, completion and request pricing in the official OpenRouter catalog. Review the access paths before building on it.
Official sourceIndependent audit
英語原文を読む →DeepSeek: DeepSeek V4 Flash 0731 (free) is now listed as a free model on OpenRouter
DeepSeek: DeepSeek V4 Flash 0731 (free) (deepseek/deepseek-v4-flash-0731:free) was detected with zero prompt, completion and request pricing in the official OpenRouter catalog. Review the access paths before building on it.
Official sourceIndependent audit
英語原文を読む →Qwen: Qwen3.8 27B (free) is now listed as a free model on OpenRouter
Qwen: Qwen3.8 27B (free) (qwen/qwen3.8-27b:free) was detected with zero prompt, completion and request pricing in the official OpenRouter catalog. Review the access paths before building on it.
Official sourceIndependent audit
英語原文を読む →Union Alpha is now listed as a free model on OpenRouter
Union Alpha (stealth/union-alpha) was detected with zero prompt, completion and request pricing in the official OpenRouter catalog. Review the access paths before building on it.
Official sourceIndependent audit
英語原文を読む →Z.ai: GLM 5.2 (free) is now listed as a free model on OpenRouter
Z.ai: GLM 5.2 (free) (z-ai/glm-5.2:free) was detected with zero prompt, completion and request pricing in the official OpenRouter catalog. Review the access paths before building on it.
Official sourceIndependent audit
英語原文を読む →inclusionAI: Ling 3.0 Flash VL (free) is now listed as a free model on OpenRouter
inclusionAI: Ling 3.0 Flash VL (free) (inclusionai/ling-3.0-flash-vl:free) was detected with zero prompt, completion and request pricing in the official OpenRouter catalog. Review the access paths before building on it.
Official sourceIndependent audit
英語原文を読む →Nex AGI adds Nex-N2.5-Mini and Nex-N2.5-Pro to OpenRouter free tier
OpenRouter has added two new models from Nex AGI, Nex-N2.5-Mini and Nex-N2.5-Pro, to its free API tier.
Official sourceIndependent audit
英語原文を読む →Nex AGI: Nex-N2.5-Pro (free) is now listed as a free model on OpenRouter
Nex AGI: Nex-N2.5-Pro (free) (nex-agi/nex-n2.5-pro:free) was detected with zero prompt, completion and request pricing in the official OpenRouter catalog. Review the access paths before building on it.
Official sourceIndependent audit
英語原文を読む →Nex AGI: Nex-N2.5-Mini (free) is now listed as a free model on OpenRouter
Nex AGI: Nex-N2.5-Mini (free) (nex-agi/nex-n2.5-mini:free) was detected with zero prompt, completion and request pricing in the official OpenRouter catalog. Review the access paths before building on it.
Official sourceIndependent audit
英語原文を読む →inclusionAI Ling 3.0 Flash Sante added to OpenRouter
OpenRouter has added the inclusionAI Ling 3.0 Flash Sante model to its API catalog. The model is available with a free tier and supports a context length of 262,144 tokens.
Official sourceIndependent audit
英語原文を読む →inclusionAI: Ling 3.0 Flash Sante (free) is now listed as a free model on OpenRouter
inclusionAI: Ling 3.0 Flash Sante (free) (inclusionai/ling-3.0-flash-sante:free) was detected with zero prompt, completion and request pricing in the official OpenRouter catalog. Review the access paths before building on it.
Official sourceIndependent audit
英語原文を読む →Groq adds Qwen/Qwen3.8-27B to supported models
Groq has added the Qwen/Qwen3.8-27B model to its list of supported preview models, with a token speed of 450 tps, a context window of 131,042 tokens, and a maximum completion length of 16,384 tokens.
Official sourceIndependent audit
英語原文を読む →Ox Alpha After the Preview: Identity, Pricing, and Retention
The Ox Alpha preview has changed: OpenRouter now identifies it as Z.AI GLM-5.3-Flash. Recheck the model route, price and data-retention terms before using it.
Official sourcesHuman checked
英語原文を読む →GPT-5.6 and Gemini 3.7 Flash: A Dated API Migration Checklist
A record of the August 2026 releases, refreshed with current official model guidance: verify model IDs, interfaces, account limits and workload tests before migrating.
Official sourcesHuman checked
英語原文を読む →Choosing an API for Image, Audio and Video Applications
Select multimodal APIs by input, output, transport and fallback requirements—not by a text-model leaderboard.
Official sourcesHuman checked
英語原文を読む →Designing an AI Agent That Can Survive a Free API Failure
Separate agent state, model routing, tool permissions and API keys so a provider outage or quota change does not stop the whole workflow.
Official sourcesHuman checked
英語原文を読む →A Model Release Is Not a Migration Plan: What to Test Before You Switch
Separate launches, previews, alias changes and shutdown notices, then migrate an AI application with a reproducible contract test.
Official sourcesHuman checked
英語原文を読む →LiquidAI LFM2.5-2.6B added to OpenRouter free tier
OpenRouter has added the LiquidAI LFM2.5-2.6B model to its free API access list, providing developers with a new zero-cost option for inference tasks.
Official sourceIndependent audit
英語原文を読む →Z.ai GLM 5.2 added to OpenRouter free tier
OpenRouter has added Z.ai GLM 5.2 as a free model option, providing developers with access to this model without cost via the OpenRouter API.
Official sourceIndependent audit
英語原文を読む →OpenRouter adds Dots Studio Dots3-Note Preview to free tier
OpenRouter has added the Dots Studio Dots3-Note Preview model to its free API access list. The model supports a context length of 512,000 tokens.
Official sourceIndependent audit
英語原文を読む →How to Compare Free AI APIs Using Evidence Instead of Model Count
Compare free AI APIs by verified access, usable limits, compatibility, failure behavior, and change risk—not the size of a model directory.
Official sourcesHuman checked
英語原文を読む →OpenAI-Compatible APIs: What to Test Before Calling Them Drop-In Compatible
A practical compatibility test for request shape, response semantics, errors, limits, and the features your application actually depends on.
Official sourcesHuman checked
英語原文を読む →A Practical Checklist for Verifying a Free LLM API Before You Build
A developer checklist for verifying free LLM API access, testing evidence, and checking operational limits before relying on a provider.
Official sourcesHuman checked
英語原文を読む →A Five-Check Smoke Test for an OpenAI-Compatible Endpoint
A compact sequence for separating interface compatibility from authentication, generation, and production-readiness claims.
Measured locallyHuman checked
英語原文を読む →Local ChatGPT2API Smoke Test: What Worked
A bounded record of one authenticated local completion—and the reliability and production questions it cannot answer.
Measured locallyHuman checked
英語原文を読む →How to create an NVIDIA NIM API key
The current official signup path, account requirements and the safe next step after the key is created.
Official pathPractical steps
英語原文を読む →Cerebras starter credits require payment verification
What the current starter-credit rule changes for users expecting a no-card free API route.
Official sourceAccess impact
英語原文を読む →Connect an OpenAI-compatible provider to Codex
Where compatibility ends, how to keep the key local and when an adapter is required.
ConfigurationKey safety
英語原文を読む →