Models & Cost
Articles on model selection, GPT-5.x capabilities, cost optimisation and pricing strategies for Codex.
95 articles
To Run or Not to Run: What 7,745 Agent Traces Reveal About the Cost-Effectiveness of Code Execution — and How to Wire Selective Testing into Codex CLI
To Run or Not to Run: What 7,745 Agent Traces Reveal About the Cost-Effectiveness of Code Execution — and How to Wire Selective Testing into Codex CLI
Mixed-Model Agent Teams: What AgentCARD Reveals About Role Bottlenecks — and How to Wire Heterogeneous Workflows in Codex CLI
AgentCARD's Shapley diagnostics show heterogeneous agent teams beat homogeneous setups by up to 44% accuracy or 12× lower cost. Here is how to replicate that advantage with Codex CLI named profiles and custom subagent definitions.
Reasoning Effort, Not Tool Access, Buys First-Try Reliability: What 90 Agent Runs Reveal — and How to Configure Codex CLI's Reasoning Levels
Reasoning Effort, Not Tool Access, Buys First-Try Reliability: What 90 Agent Runs Reveal — and How to Configure Codex CLI’s Reasoning Levels
Specialize Roles, Mix Deployments: What AgentCARD Reveals About Heterogeneous Agent Teams — and How Codex CLI's Custom Agent Definitions Deliver It
Specialize Roles, Mix Deployments: What AgentCARD Reveals About Heterogeneous Agent Teams — and How Codex CLI’s Custom Agent Definitions Deliver It
Where Do Your Tokens Go? What Empirical Research Reveals About Coding Agent Token Consumption — and How to Control It with Codex CLI
Where Do Your Tokens Go? What Empirical Research Reveals About Coding Agent Token Consumption — and How to Control It with Codex CLI
GPT-5.6 Sol, Terra, and Luna: What OpenAI's Three-Tier Model Family Means for Codex CLI Workflows
GPT-5.6 Sol, Terra, and Luna: What OpenAI’s Three-Tier Model Family Means for Codex CLI Workflows
Agent-as-a-Router: What the First Agentic Model Routing Benchmark Means for Codex CLI Multi-Model Workflows
Agent-as-a-Router: What the First Agentic Model Routing Benchmark Means for Codex CLI Multi-Model Workflows
GPT-5.6 Sol, Terra, and Luna: What the Three-Tier Model Preview Means for Codex CLI Developers
GPT-5.6 Sol, Terra, and Luna: What the Three-Tier Model Preview Means for Codex CLI Developers
Early Termination Strategies for Coding Agents: What EET and AgentStop Mean for Codex CLI Cost Efficiency
Early Termination Strategies for Coding Agents: What EET and AgentStop Mean for Codex CLI Cost Efficiency
MiniMax M3: What the First Open-Weight Model to Beat GPT-5.5 on SWE-Bench Pro Means for Codex CLI Model Routing
MiniMax M3: What the First Open-Weight Model to Beat GPT-5.5 on SWE-Bench Pro Means for Codex CLI Model Routing
Don't Break the Cache: What the Prompt Caching Research Means for Codex CLI Cost and Latency Optimisation
Don’t Break the Cache: What the Prompt Caching Research Means for Codex CLI Cost and Latency Optimisation
Context Pruning for Coding Agents: What SWE-Pruner, Pichay, and ContextBudget Mean for Codex CLI Token Management
Context Pruning for Coding Agents: What SWE-Pruner, Pichay, and ContextBudget Mean for Codex CLI Token Management
Do Programming Languages Still Matter? What the Chess Engine Polyglot Study Means for Codex CLI Language Selection and Cost Strategy
A polyglot study built 34 chess engines in 17 languages using Codex CLI and Claude Code. The results refute 'language doesn't matter' and reveal concrete cost and performance implications for Codex CLI practitioners.
SkillReducer: What the First Large-Scale Skill Bloat Study Means for Codex CLI Token Efficiency
SkillReducer: What the First Large-Scale Skill Bloat Study Means for Codex CLI Token Efficiency
Benchmark Literacy: A Practitioner's Guide to Reading Coding Agent Benchmarks Critically
Benchmark Literacy: A Practitioner’s Guide to Reading Coding Agent Benchmarks Critically
Frontier Agents and Metaprogramming: What EsoLang-Bench Reveals About Codex CLI Reasoning Effort, Tool Budgets, and Strategy Transfer
Frontier Agents and Metaprogramming: What EsoLang-Bench Reveals About Codex CLI Reasoning Effort, Tool Budgets, and Strategy Transfer
When the Harness Outweighs the Model: What Claw-SWE-Bench, Harness-Bench, and UTBoost Mean for Codex CLI Configuration Strategy
Three recent papers independently prove that agent harness design is at least as important as model selection. This article maps their findings to practical Codex CLI configuration patterns.
KiloBench and the Cost-per-Task Revolution: What Harness-Aware Efficiency Benchmarks Mean for Codex CLI Model Selection
KiloBench and the Cost-per-Task Revolution: What Harness-Aware Efficiency Benchmarks Mean for Codex CLI Model Selection
Five Production Configuration Patterns for Codex CLI in the Post-Subsidy Era: Billing Resilience, Model Routing, and Automation Hygiene for June 2026
Five Production Configuration Patterns for Codex CLI in the Post-Subsidy Era
Codex CLI Reasoning Tiers: Mapping the June 2026 Model Picker to CLI Profiles for Cross-Surface Consistency
Codex CLI Reasoning Tiers: Mapping the June 2026 Model Picker to CLI Profiles for Cross-Surface Consistency
The GPT-5.3-Codex Countdown: Migrating Your Codex CLI Configuration Before the June 30 API Deadline
The GPT-5.3-Codex Countdown: Migrating Your Codex CLI Configuration Before the June 30 API Deadline
GPT-5-Codex Refreshed: The June 14 Model Update and the Mid-2026 Model Selection Decision Tree for Codex CLI
GPT-5-Codex Refreshed: The June 14 Model Update and the Mid-2026 Model Selection Decision Tree for Codex CLI
The Agent Billing Convergence: Managing Developer Costs as Copilot, Codex, and Claude Code All Move to Usage-Based Pricing
The Agent Billing Convergence: Managing Developer Costs as Copilot, Codex, and Claude Code All Move to Usage-Based Pricing
Workspace Agents Credit Pricing Starts July 6: A Codex CLI Practitioner's Budget Preparation Guide
Workspace Agents Credit Pricing Starts July 6: A Codex CLI Practitioner’s Budget Preparation Guide
Beyond Model Chasing: Why the June 2026 Benchmark Convergence Means Your Codex CLI Configuration Is the Real Competitive Advantage
Beyond Model Chasing: Why the June 2026 Benchmark Convergence Means Your Codex CLI Configuration Is the Real Competitive Advantage
The Silent Model Downgrade Problem: Detecting and Defending Against GPT-5.5 Quality Regression in Codex CLI Workflows
The Silent Model Downgrade Problem: Detecting and Defending Against GPT-5.5 Quality Regression in Codex CLI Workflows
The AI Token Price War: OpenAI's Pre-IPO Price Cuts, the SpaceX Nasdaq Debut, and What Codex CLI Developers Should Budget For
The AI Token Price War: OpenAI’s Pre-IPO Price Cuts, the SpaceX Nasdaq Debut, and What Codex CLI Developers Should Budget For
Claude Fable 5 Enters the Arena: What Anthropic's Benchmark-Topping Model Means for Codex CLI Multi-Model Strategies
Claude Fable 5 Enters the Arena: What Anthropic’s Benchmark-Topping Model Means for Codex CLI Multi-Model Strategies
Codex CLI Rate-Limit Reset Banking and Usage Optimisation: Managing Agent Costs from the Terminal
Codex CLI Rate-Limit Reset Banking and Usage Optimisation: Managing Agent Costs from the Terminal
Terminal-Bench 2.1 and the June 2026 Benchmark Landscape: Why the Harness Matters More Than the Model for Codex CLI Developers
Terminal-Bench 2.1 and the June 2026 Benchmark Landscape: Why the Harness Matters More Than the Model for Codex CLI Developers
OpenAI's S-1 Filing: What the IPO Path Means for Codex CLI Developers
OpenAI’s S-1 Filing: What the IPO Path Means for Codex CLI Developers
Diagnosing and Reducing Codex CLI Token Consumption: A Practitioner's Toolkit for the June 2026 Quota Landscape
Diagnosing and Reducing Codex CLI Token Consumption: A Practitioner’s Toolkit for the June 2026 Quota Landscape
From One Year to Two Weeks: How Wasmer Built Edge.js with Codex CLI and What Systems Programmers Should Learn from It
From One Year to Two Weeks: How Wasmer Built Edge.js with Codex CLI and What Systems Programmers Should Learn from It
The MCP Tax: When Shell Commands Beat MCP Servers in Codex CLI Workflows
The MCP Tax: When Shell Commands Beat MCP Servers in Codex CLI Workflows
Multi-Model Daily Workflows with Codex CLI: Routing GPT-5.5, Spark, Mini, and Open-Weight Models for Cost, Speed, and Quality
Multi-Model Daily Workflows with Codex CLI: Routing GPT-5.5, Spark, Mini, and Open-Weight Models for Cost, Speed, and Quality
Coding Agent Landscape, June 2026: How Codex CLI v0.137 Stacks Up Against Copilot Flex, Devin Desktop, Antigravity 2.0, and Kiro
Coding Agent Landscape, June 2026: How Codex CLI v0.137 Stacks Up Against Copilot Flex, Devin Desktop, Antigravity 2.0, and Kiro
Codex CLI After the Pro Boost: Rate Limit Reality, Token Economics, and Cost Optimisation for June 2026
Codex CLI After the Pro Boost: Rate Limit Reality, Token Economics, and Cost Optimisation for June 2026
The Codex Model Sunset: June–July 2026 Deprecation Timeline, Migration Paths, and Config Recipes
The Codex Model Sunset: June–July 2026 Deprecation Timeline, Migration Paths, and Config Recipes
The June 2026 Coding Agent Billing Reset: What Four Simultaneous Pricing Changes Mean for Your Codex CLI Budget
The June 2026 Coding Agent Billing Reset: What Four Simultaneous Pricing Changes Mean for Your Codex CLI Budget
The June–July 2026 Model Sunset: Migrating from GPT-5.2 and GPT-5.2-Codex in Your Codex CLI Workflows
The June–July 2026 Model Sunset: Migrating from GPT-5.2 and GPT-5.2-Codex in Your Codex CLI Workflows
Planning for Token Meltdown: How to Route Local to Paid Automatically
When model providers stop subsidising token costs, your bill explodes overnight. A routing layer that tries local models first and promotes to cloud only when necessary is how you survive the transition. LiteLLM running as a local proxy gives you automatic fallback based on failures, latency, or context window limits.
Gemini 3.5 Flash vs GPT-5.5 and codex-mini: Coding Model Benchmark Comparison After Google I/O 2026
Google I/O 2026 dropped Gemini 3.5 Flash on 19 May with a bold claim: it beats Gemini 3.1 Pro on coding benchmarks whilst running four times faster than.
Grok Build Enters the Ring: How xAI's Parallel-Agent CLI Compares to Codex CLI
On 14 May 2026, Elon Musk posted a broad call for beta testers of Grok Build, xAI's first terminal-native coding agent. The tool enters a market dominated.
The Agentic Pricing Wars: OpenAI's Enterprise Migration Offer, Anthropic's Quota Expansion, and the Real Cost of Switching
On 13 May 2026, OpenAI launched a 30-day Switch to Codex promotion offering two months of free enterprise usage for organisations transitioning from.
GPT-5.3-Codex Deep Dive: Benchmarks, CLI Configuration, and Interactive Coding Workflows
GPT-5.3-Codex landed on 5 February 2026 as OpenAI's flagship coding model, promising industry-leading agentic performance alongside a 25 % speed improvement.
Codex CLI Context Compaction Under GPT-5.5: Diagnosing Failures, Configuring Fallbacks, and Keeping Long Sessions Alive
Since GPT-5.5 became the default model in Codex CLI, a wave of compaction failures has disrupted long-running sessions for practitioners worldwide. GitHub.
Prompting GPT-5.5 in Codex CLI: Outcome-First Instructions, AGENTS.md Patterns, and Reasoning Effort Tuning
GPT-5.5 landed in Codex CLI in late April 2026 as OpenAI's newest frontier model, bringing stronger planning, tool use, and multi-step follow-through.
Codex CLI Model Routing in May 2026: GPT-5.5, GPT-5.4, Codex-Spark, and When to Use Each
The model lineup available to Codex CLI developers has changed materially since March 2026. GPT-5.5 arrived in late April as the new frontier, GPT-5.4.
GPT-5.5 Instant and chat-latest: Dynamic Model Pointers for Codex CLI Developers
On 5 May 2026, OpenAI replaced GPT-5.3 Instant with GPT-5.5 Instant as the default ChatGPT model and simultaneously shipped a new API model alias.
Terminal Agent Showdown: Codex CLI vs Claude Code vs Gemini CLI in May 2026
The terminal agent race has intensified since the three-way contest crystallised in late 2025. OpenAI's Codex CLI (v0.128.0, Rust-native), Anthropic's.
The Codex CLI Model Landscape in May 2026: A Practitioner's Routing Guide
Three months ago, Codex CLI users chose between GPT-5-Codex and its Spark variant. Today the model picker lists five production options — six if you count.
Do Agent-Written Tests Actually Help? What Six LLMs on SWE-bench Reveal and How to Rethink Your Codex CLI Testing Strategy
The instinct to make coding agents write tests is strong — and understandable. Test-driven development has been a pillar of professional software.
Codex CLI Daily Driver Setup for May 2026: An Opinionated Configuration Guide
Codex CLI v0.128 is the most configurable release yet. Between named profiles, persistent memories, configurable keymaps, goal workflows.
Codex CLI Named Profiles: A Cookbook of Ready-to-Use Configuration Templates
Named profiles are one of the most underused features in Codex CLI. Instead of juggling CLI flags or maintaining separate config files.
Codex CLI Service Tiers Explained: Flex, Standard, and Fast Mode for Cost and Speed Optimisation
Every codex exec invocation and every interactive session burns tokens. Whether you are running a quick lint fix or a six-hour codebase migration.
GitHub Copilot's Usage-Based Billing Shift: What It Means for Codex CLI Teams
On 27 April 2026, GitHub announced that every Copilot plan will move from premium request units to usage-based billing with GitHub AI Credits on 1 June 2026.
Codex at Four Million: What Three Weeks of Hypergrowth Reveals About the Agentic Coding Market
On 28 April 2026, the OpenAI-AWS partnership announcement casually confirmed that more than 4 million people now use Codex every week .
Architecture Decision Records with Codex CLI: Automated ADR Generation, Governance, and the Agent-Architecture Gap
Every team says they will write Architecture Decision Records. Few actually do. The friction is well understood.
Codex CLI for Dockerfile Optimisation: Multi-Stage Builds, Layer Caching, and Security Hardening
Dockerfiles look simple. They are deceptively hard to get right. A naively written Dockerfile for a Node.js application can produce a 1.2 GB image with a.
The Coding Agent CLI Landscape in Late April 2026: GPT-5.5, Five-Way Competition, and What Changed This Month
Two weeks ago, the coding agent CLI market was a three-horse race. Today it is five — and the dynamics have shifted more in April 2026 than in any month.
JavaScript-to-TypeScript Migration with Codex CLI: Gradual Typing Strategies for Large Codebases
Migrating a JavaScript codebase to TypeScript remains one of the most requested — and most dreaded — modernisation tasks in 2026.
Codex CLI Cost Calculator: Building a Token Budget Estimator for Mixed-Model Workflows
With OpenAI's April 2026 shift to token-based billing for Codex, understanding exactly where your credits go across models like o4-mini, gpt-5.4.
Debugging with Codex CLI: Systematic Bug-Hunting Patterns for GPT-5.5
Debugging is one of the highest-leverage uses of Codex CLI, yet most practitioners treat it as an afterthought.
GPT-5.5's Million-Token Context Window: Practical Strategies for Codex CLI Long-Context Workflows
GPT-5.5 shipped on 23 April 2026 with a headline that most coverage buried beneath benchmark tables: the API context window doubles from 512K to 1M tokens.
Workspace Agents and Codex Slack Integration: From CLI Automations to Team-Shared Agentic Workflows
On 22 April 2026, OpenAI launched workspace agents — shared, persistent agents powered by Codex that run in the cloud and integrate directly into Slack.
The Codex CLI Speed Stack: Fast Mode, Reasoning Effort, Spark, and Performance Tuning
Codex CLI now ships four independent speed levers, each with its own trade-off envelope. This article maps every lever — Fast service tier, reasoning.
The Codex Subscription API: Programmatic Access to GPT-5.5 Through Your ChatGPT Plan
When OpenAI launched GPT-5.5 on 23 April 2026, a curious limitation accompanied the announcement: the model is available only through ChatGPT subscription.
DeepSeek V4 as a Codex CLI Provider: Frontier-Class Coding at a Fraction of the Cost
DeepSeek V4 landed today — 24 April 2026 — and the numbers deserve attention. V4-Pro scores 80.6% on SWE-bench Verified while charging $3.48 per million.
GPT-5.5 Migration Cookbook: Effort Tuning, Cost Comparison, Prompt Adjustments
GPT-5.5 — codenamed Spud — landed on 23 April 2026, less than two months after GPT-5.4. It is already the recommended model for most Codex tasks.
GPT-5.5 Drops: What Changes for Codex Users
Six weeks. That is the gap between GPT-5.4 and GPT-5.5. OpenAI released its newest frontier model on 23 April 2026, rolling it out simultaneously to ChatGPT.
MCP Schema Bloat and System Prompt Tax: Performance Impact of Tool Definitions
Every MCP server you connect to Codex CLI injects its full tool manifest — JSON schemas with parameter descriptions, type annotations, enum constraints.
Prompt Caching in Codex CLI: How the Agent Loop Stays Linear and How to Maximise Cache Hits
Every Codex CLI session resends the full conversation history on each turn. Without mitigation, this is quadratic in cost and latency. The engineering.
The Deep Researcher Pattern: Building 24/7 Autonomous Experimentation Loops with Codex CLI
A new open-source framework called Deep Researcher Agent, published by Xiangyue Zhang at the University of Tokyo in April 2026.
When to Use Multi-Agent vs Single-Agent: A Practical Framework for Codex CLI Teams
Codex CLI's subagent system lets you spawn parallel agents for concurrent work — but more agents does not always mean better results. Recent academic.
Purpose-Built Agent Models: What codex-auto-review Tells Us About the Future of Specialised AI
On 16 April 2026, a single-commit pull request landed in the Codex CLI repository that carries outsized strategic significance.
Codex as a GitHub Coding Agent: Agent HQ, Model Selection, and Cloud-Based Code Review
Most coverage of Codex focuses on the CLI — the open-source terminal agent you install with npm install -g @openai/codex.
The April 2026 Model Deprecation Wave: Migrating Your Codex CLI Configuration
On 14 April 2026, OpenAI completed the largest model retirement in Codex CLI's history. Six models — gpt-5.2-codex, gpt-5.1-codex-mini, gpt-5.1-codex-max.
GitHub Agent HQ Model Selection: Choosing GPT-5.4 vs GPT-5.3-Codex, Custom Agents, and the Multi-Agent GitHub Workflow
GitHub's Agent HQ — the platform that lets developers assign coding tasks to Copilot, Claude, and Codex directly from issues, pull requests, and the Agents.
What the ETH Zurich Paper Gets Wrong (and Right) About AGENTS.md
In February 2026, researchers at ETH Zurich published a paper that sent shockwaves through the AI-assisted development community: Evaluating AGENTS.md.
Dynamic Model Routing in Codex CLI: Mid-Session Switching, /fast Mode, and Service Tier Workflows
Not every turn in a Codex CLI session demands the same model, the same speed, or the same reasoning depth. A planning pass benefits from deep deliberation.
The Three-CLI Toolkit: Running Codex CLI, Claude Code, and Gemini CLI as a Unified Development Stack
The terminal AI coding agent market has consolidated around three big-lab native tools: OpenAIs Codex CLI, Anthropics Claude Code.
The Complete Codex CLI Pricing Guide: Subscriptions, Tokens, Cost Optimisation, and Competitive Analysis
Every AI coding agent runs on tokens, and tokens cost money. Codex CLI is unique among major AI coding agents: it is open-source, runs locally, and supports.
Context Compaction Showdown: How Codex CLI, Claude Code, and 5 Other Agents Handle Full Context Windows
Every AI coding agent eventually fills its context window. What happens next — the compaction strategy — determines whether your session gracefully.
Codex Pay-As-You-Go Pricing: Modelling Costs for Multi-Agent Workflows
On 2 April 2026, OpenAI replaced Codex's per-message credit system with token-based billing aligned to API usage. The same announcement lowered the annual.
Codex CLI 3 Million Users: Growth Trajectory and What the Usage Limit Reset Strategy Means
On 8 April 2026, Sam Altman announced that Codex had crossed three million weekly active users.
Codex CLI Custom Model Providers: Azure, Vercel, Local LLMs and Dynamic Bearer Tokens
Codex CLI ships wired to OpenAI's hosted models, but the [model_providers] configuration table lets you point it at any OpenAI-compatible endpoint — Azure.
GPT-5.3-Codex-Spark and the Cerebras Inference Stack: Real-Time Coding at 1,000 Tokens per Second
GPT-5.3-Codex-Spark is OpenAI's first model purpose-built for real-time coding iteration, and the first production model served entirely on non-NVIDIA.
GPT-5.4 Computer Use and Tool Search in Codex CLI: Visual Debugging, Deferred Loading, and /fast Mode
GPT-5.4, released on 5 March 2026, is OpenAI's first mainline reasoning model to ship with native computer-use capabilities alongside the frontier coding.
gpt-5-codex: The New Codex Flagship and What It Means for Your Workflow
In late March 2026, OpenAI shipped two new models — gpt-5-codex and gpt-5-codex-mini — announced by @thsottiaux.
GPT-5.4 mini in Codex CLI: Subagent Delegation, Model Routing and the Tiered Inference Architecture
OpenAI released GPT-5.4 mini and GPT-5.4 nano on 17 March 2026, and they represent something more significant than two incremental model updates.
Designing Workflows for Codex-Spark: When Inference Becomes Near-Instant
At 65–70 tokens per second, a 400-token code response takes roughly 6 seconds — a noticeable pause.
Codex CLI Cost Management: Token Strategy, Model Routing and Quota Control
The biggest surprise in Codex deployments isn't the cost of output tokens — it's the accumulated cost of conversation history.
GPT-5.3-Codex-Spark: The Cerebras-Powered Ultra-Fast Coding Model
On 14 January 2026, OpenAI announced a multi-year partnership with Cerebras Systems. Four weeks later, on 12 February 2026, the first concrete output.
Reasoning Effort Tuning: Minimal to xhigh for Cost and Speed
Codex CLI's reasoning engine has a single knob that dramatically affects cost, speed, and quality: model_reasoning_effort.
Model Selection in Codex CLI: Current Models and When to Use Each
Codex CLI exposes model selection as a first-class concern. You can specify a model on the command line for a single invocation.