Blog
Read about our latest announcements.DeepSeek V4 Pro vs Flash: A Buyer's Decision Guide
Compare DeepSeek V4 Pro and Flash across benchmarks, reasoning, coding, speed, context length, and API pricing to choose the right model.
DeepSeek V4 Flash Benchmarks: Scores, Speed, Pricing and How It Compares
Explore DeepSeek V4 Flash benchmarks, SWE-bench and GPQA scores, API pricing, inference speed, hardware requirements, and comparisons with leading models.
DeepSeek V4 Flash API Pricing: Rates, Cost Calculator and Provider Comparison
Compare DeepSeek V4 Flash API pricing, cache rates, provider costs, free access, and worked cost examples for chat, coding, and reasoning workloads.
Introducing Organization: A Better Way for Teams to Build on ZenMux
Manage team AI usage with ZenMux Organizations: individual API keys, a shared PAYG balance, member limits, centralized billing, logs, and usage visibility.
Why Does an AI Model Say It Is Another Model?
ZenMux asked 27 LLMs “Who are you?” 29,700 times to explore why AI models sometimes misidentify themselves — and why self-identification is not reliable proof of model routing.
Ling-3.0-flash API Now Available on ZenMux: Built for Agentic Coding
Use the Ling-3.0-flash API on ZenMux for agentic coding, long-horizon tasks, and tool workflows, with 256K context and a practical API example.
Token Economics: What Happens When We Price Every "Eastern Model" to Match DeepSeek?
Token Economics: What Happens When We Price Every "Eastern Model" to Match DeepSeek?
ZenMux Token Economics: When Model Prices Meet, What Decides the Winner?
ZenMux Token Economics brings 10+ popular AI models to DeepSeek-level pricing. Test them on real work and let real usage decide.
Claude Agent SDK with ZenMux: Cut Agent Loop Costs, Keep Loops Running
Point the Claude Agent SDK at ZenMux to access 200+ models, route phases of your agent loop to cheaper models, and keep loops running through provider outages — all behind one API key.
We Spent 1,000 Dollars Asking 27 Frontier LLMs From 16 Vendors One Question 29,700 Times: "Who Are You?"
We spent about 1,000 dollars asking 27 frontier LLMs from 16 vendors the same question 29,700 times: “Who are you?” Most of the time they knew. But in 7.1% of answers they claimed another vendor’s identity—mostly Anthropic, OpenAI, or Google—and that confusion spiked with language and de-branding prompts. This isn’t proof of distillation; it’s an identity echo in the training data. Full data and tools are open source.
An Experimental Validation Plan for the Feedback That "ZenMux Has Worse Cache Hit Rates Than OpenRouter"
This article describes an experimental framework for comparing cache hit rates between ZenMux and OpenRouter. By fixing the model, provider, and prompt sequence, and by measuring cached_tokens and token_hit_rate across multi-turn conversations, the experiment shows that there is no significant difference in cache hit rate between the two platforms under the tested conditions.
Debugging a Claude Code Freeze
Recently, several users reported that Claude Code would freeze when calling models through ZenMux — after sending a prompt, nothing appeared on screen, and the only option was to force quit. As a model aggregation platform, we needed to figure out whether this was a ZenMux issue or something else entirely.