Latest AI News
Latest reporting and analysis from Kingy AI on AI launches, tools, model releases, research, funding, policy, safety, and business moves.
-
The cheapest AI API models in 2026—priced as workloads, not headlines
One million input plus 250,000 output tokens, before quality and operational costs. The cheapest accepted row in the first Kingy.ai AI API Price Index…
-
OpenAI vs Anthropic API pricing in 2026: the workload decides
OpenAI and Anthropic can land surprisingly close on standard token cost, but the similarity disappears when context bands, batch and fast service are mixed…
-
AI API prices are cheap—until the label hides a different product
Output is commonly the expensive side of the API bill. AI companies publish plenty of prices. They do not publish one clean market. The…
-
Independent Open-Source Coding Agent Field Guide
A source-audited guide to OpenCode, Codex CLI, OpenHands, Cline, goose and Qwen Code, with exact product boundaries, local smoke evidence and no fabricated leaderboard.
-
Is an Open-Source AI Ban Coming? The Real Risk Is Narrower—and More Plausible
No blanket open-source AI ban is imminent. The real risk is a capability-triggered restriction on frontier weights, with Chinese services targeted first.
-
OpenCode + Local LLMs: Build, Test, and Safely Operate a Local AI Coding Agent
A tested OpenCode 1 and Ollama guide covering local-model setup, hardware, permissions, privacy verification, failure modes and local-versus-cloud benchmarking.
-
How Harvey Turned Kimi K3 Into Tenet—and Rewrote the Case for Open-Weight AI
Harvey turned Kimi K3 into Tenet with legal environments, expert data and reinforcement learning. The results—and caveats—reshape the open-weight debate.
-
Ox Alpha Is Free for a Week: Inside OpenCode’s Mystery 1M-Context Coding Model
OpenCode has made the anonymous Ox Alpha coding model free for one week. We separate the confirmed 1M-context and privacy claims from speculation, explain…
-
Is Ox Alpha Really GLM-5.3 Flash? The Evidence So Far
Ox Alpha looks increasingly like a GLM-5.x-family model. Tokenizer matches and a GLM-5.3-specific reasoning error strengthen the case, but no official source has confirmed…
-
Coding-Agent Prompt-Injection Benchmarks: A Practical Crosswalk
Seven prompt-injection benchmarks measure different targets. This crosswalk shows which scores can—and cannot—be compared.