Breaking

Open-Source 'opencodex' Setup Lets Developers Run Fable 5, Kimi K3 and Grok 4.5 Inside OpenAI's Codex

July 21, 2026 at 06:45 EDT

Software Dev & Coding
Foundation Models
Open Source

A newly circulated open-source configuration allows developers to route three of the year's frontier models — Anthropic's Fable 5, Moonshot AI's Kimi K3 and xAI's Grok 4.5 — directly inside OpenAI's Codex coding agent through a single setup. The approach, built around a repository informally referred to as "opencodex," pairs OpenAI's terminal-based Codex CLI with the APIs of competing model providers so that a user can call any of the three models from one workflow rather than switching between vendor-specific tools.

Mid-2026 · Vendor-Agnostic Coding Agents

Three frontier models, one Codex terminal

An open-source "opencodex" setup wires Anthropic's Fable 5, Moonshot's Kimi K3 and xAI's Grok 4.5 into OpenAI's Codex CLI — a single install that routes each task to whichever model fits best.

3 → 1

Three vendor APIs consolidated into one Codex workflow

Token context window reported for Kimi K3

~4.7×

Cost gap for one task: $2.49 vs $11.80

Cost per representative task

Why price, not benchmarks, may decide everyday use.

$2.49

Grok Build

$11.80

Claude Code

Benchmarks: Fable 5 (max) vs Grok 4.5

SWE Bench Pro

80.4% Fable 5

64.7% Grok 4.5

Terminal Bench 2.1

84.3% Fable 5

83.3% Grok 4.5

DeepSWE 1.1

70% Fable 5

53% Grok 4.5

A single routing layer, split by strength

Fable 5
Plan & code review

→

Kimi K3
Front-end, 1M context

→

Grok 4.5
Search & knowledge tasks

The Draw

Mix model strengths in one harness
One install, no vendor tool-switching
Route to the cheapest capable model

The Cautions

Costs and rate limits stack up fast
Needs valid API access at all 3 vendors
Routing layer is community-built, unsupported

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.