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。
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。
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.
A new anonymous reasoning model, Ox Alpha (stealth/ox-alpha), is available free through OpenRouter and Nous Portal with a 1M-token context window. Here is what is documented, what is not, and how to verify before building.
Two current model updates emphasize coding and agent workflows, but migration decisions still depend on model IDs, interfaces, limits and representative tests.
Select multimodal APIs by input, output, transport and fallback requirements—not by a text-model leaderboard.
Separate agent state, model routing, tool permissions and API keys so a provider outage or quota change does not stop the whole workflow.
Separate launches, previews, alias changes and shutdown notices, then migrate an AI application with a reproducible contract test.
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.
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.
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.
Compare free AI APIs by verified access, usable limits, compatibility, failure behavior, and change risk—not the size of a model directory.
A practical compatibility test for request shape, response semantics, errors, limits, and the features your application actually depends on.
A developer checklist for verifying free LLM API access, testing evidence, and checking operational limits before relying on a provider.
A compact sequence for separating interface compatibility from authentication, generation, and production-readiness claims.
A bounded record of one authenticated local completion—and the reliability and production questions it cannot answer.
The current official signup path, account requirements and the safe next step after the key is created.
What the current starter-credit rule changes for users expecting a no-card free API route.
Where compatibility ends, how to keep the key local and when an adapter is required.