Blog
Engineering deep-dives, product updates and AI infrastructure notes from the OrcaRouter team.
Guides & InsightsFeaturedTencent Hy4 vs Tencent Hy4 Preview: The Newer Model Is the Less Open One
Tencent Hy4 adds vision and computer use at the preview's exact price, but its weights are still a promise. What changed and which to pick.
Guides & InsightsTencent Hy4 vs Tencent HY3: Four Times the Context, Four and a Half Times the Price
Tencent Hy4 costs 4.6x more per input token than Tencent HY3 and 1.9x on a cached agent session. Specs, prices, and which one to run.
Guides & InsightsTencent Hy4 Goes GA With Vision and Computer Use — and No Benchmark Table
Tencent Hy4 shipped on 4 November with image understanding, computer use and unchanged pricing — but no public benchmark table and no weights yet.
Guides & InsightsClaude Opus 5.5 vs Gemini 4 Carbon: A 12% Ceiling and a Third of the Bill
SemiAnalysis's GDPx puts Claude Opus 5.5 first at 12% and Gemini 4 Carbon second at a third of the cost. What the numbers settle, and what they don't.
Guides & Insightspplx-decider-flash vs pplx-decider: The 4B Is Cheaper Until You Ask Three Questions
Flash costs half as much per token and more per request once you ask three questions. Where the 4B wins, where the 26B wins, and the exact crossover.
Guides & Insightspplx-decider-flash vs Laya: Two Kinds of Decision Model, One Wrong Comparison
A 4B multimodal judge against a 400M encoder: Laya needs fine-tuning, Flash needs verification. Latency, calibration and the benchmark nobody shares.
Guides & Insightspplx-decider-flash: Perplexity's Quiet 4B Judge, Read From the Repo
Perplexity's 4B decision model shipped on 15 October with a repo, a price and few answers. What the card proves, what it doesn't, and the billing trap.
Guides & InsightsCrouton Alpha: A Free 1M-Token Model Nobody Will Put a Name On
Crouton Alpha: a free 1M-token anonymous preview with no vendor and no benchmarks — and a retention clause worth reading before you point an agent at it.
Guides & InsightsMistral Large 4 vs Mistral Large 3: The Upgrade That Costs 40% More
Mistral Large 3 was deprecated on 31 October 2026 and its replacement costs 40% more per token. What improved, what the licence lost, and how to migrate.
Guides & InsightsMistral Large 4 vs Llama 4 Maverick: A Generation Gap, Not a Price Gap
Mistral Large 4 is 19 months newer than Llama 4 Maverick, yet only 1.3x the median price. Where the real differences are, and where 1M context breaks.
