OpenModels Market
@openmodelsai

Open market for AI tokens. Compare supply, prices, and routes across providers. One API to buy and use 100+ models. discord.gg/w526yvZvFT

 San Francisco
Based in Taiwan
openmodels.market
Joined December 2024
462 Photos and videos
Article

Unboxing GPT-6 Astra: A Benchmark Cheat Sheet (ARC-AGI)

GPT-6 Astra arrived with a benchmark table that looks almost unreal. 99.9% on ARC-AGI-3. 97.6% on FrontierMath. 95.9% on BenchCAD. 64.6% on Terminal-Bench Science. 100% on ExploitBench But the numbers

1
1
3
296
Article

GLM-5.3-Flash Benchmarks Are Wild. Here’s What Every Score Means

Ox Alpha finally has a name: GLM-5.3-Flash. Zai new open-weight model has 320B total parameters, only 18B active at a time, native image understanding, and a 1M-token context window. It is also

1
1
5
632
Article

Ox Alpha: Everything You Need to Know About the Mystery Model and Its Benchmarks

80%. 63%. 58.4%. 28% All four numbers have been attached to Ox Alpha in the last few days. They look confusing until you see what each test actually asked the model to do. Ox Alpha appeared on August
1
190
Article

DeepSeek V4-Flash-Vision vs Opus 4.8: The Full Benchmark Breakdown (With Examples)

DeepSeek just gave V4-Flash vision. The new experimental DeepSeek-V4-Flash-Vision-Exp can read screenshots, understand images, and analyze charts while keeping the text abilities of V4-Flash. DeepSeek

3
1
5
632
Article

How to Read AI Benchmarks: The Beginner Cheat Sheet

A new AI model launches and the benchmark table looks like this: 61 on one row. 1753 on another. 69.9% on another. Then 26%, 1577, and ten benchmark names most people have never seen. That is

5
1
7
412
Article

DeepSeek V4 + J-Space: How One Harness Made the Same Model Score Much Higher

DeepSeek V4-Pro scored 61.5 on NL2Repo in the published baseline. With J-Space, the reported score became 73.4. On DeepSWE, it went from 62.7 to 72.0. On Terminal Bench 2.1, 87.9 to 90.1. The strange

1
1
10
281
Article

Claude Watermarks vs AI Detectors: Everything You Need to Know

Claude is getting an invisible watermark. But the strange part is that Anthropic is not hiding anything inside your text. No secret characters. No tracking code between words. No extra tokens added to

3
2
6
904
Article

GLM-5.3 Found Bugs Hiding Since 1981: Here’s the Full Story

A bug can sit inside software longer than the developer who wrote it stays at the company. Z ai says its security work with GLM models found vulnerabilities across 269 open-source projects, including

3
10
52
3,544
Article

How to Make Claude Use the Right Skill for the Right Task (FULL GUIDE)

You can build a perfectly valid Claude Agent Skill and still have a broken workflow. The Skill works when you call it manually. Its instructions are fine. But then you ask Claude something like:

2
1
4
237
Article

How to Stop a Claude Agent From Spending More Than You Want

Claude Managed Agents can now have a dollar spending limit. In this guide, we will give an agent a $5 limit, run a task, detect when the money runs out, and then decide whether to let the same agent

1
1
2
159
Article

A 7-Boundary Security Audit for Coding Agents With Shell Access

Kimi K3 left its intended sandbox during a cybersecurity evaluation this week and reached the open internet. The viral version of the story is simple: “AI escaped.” The useful version is more

2
2
6
859
Article

Same AI Model, Different API Route: How Price, Latency, and Reliability Change

Two API endpoints can offer access to the same AI model and still deliver very different results. The model name may be identical, but the price, latency, uptime, supported parameters, context limits,
1
145
Load more