Codex Pricing and Limits (Nov 2026): GPT-6.1 and GPT-5.6 Credits, Weekly Limits, API Costs
How Codex bills GPT-6.1 Sol, GPT-6.1 Terra, GPT-6 and GPT-5.6 in credits, plan prices, five-hour and weekly limits, Fast and Ultrafast multipliers, the GPT-5.5 retirement, and what the same task costs with an API key.
Codex can be paid for in two ways, and they are billed differently:
- With a ChatGPT plan (Free, Go, Plus, Pro, Business, Enterprise, Edu), Codex converts token usage into credits and draws from the plan’s included usage first. Plus and Pro users can buy extra credits when they run out.
- With an API key (CLI, SDK or IDE extension), usage bills at normal OpenAI API token prices, without cloud features.
Everything below was checked against OpenAI’s Codex pricing page and API pricing page on November 16, 2026.
Codex Credit Rate Card
Credits per 1M tokens:
| Model | Input | Cached input | Output |
|---|---|---|---|
| GPT-6 Astra | 250 | 25 | 1,250 |
| GPT-6.1 Sol | 50 | 2.5 | 250 |
| GPT-6.1 Terra | 25 | 1.25 | 125 |
| GPT-6 Sol | 50 | 5 | 250 |
| GPT-6 Luna | 2.5 | 0.25 | 12.5 |
| GPT-5.6 Sol | 100 | 10 | 500 |
| GPT-5.6 Terra | 50 | 5 | 300 |
| GPT-5.6 Luna | 5 | 0.5 | 30 |
Codex has no separate cache-write charge. OpenAI says a typical GPT-5.6 Sol task uses 2 to 15 credits.
These rates are lower than when GPT-5.6 launched in July, when Sol cost 125 / 12.5 / 750 credits, Terra 62.5 / 6.25 / 375 and Luna 25 / 2.5 / 150. If you planned Codex usage from older numbers, GPT-5.6 Luna now uses one fifth of the credits it did.
What one task costs in credits
Take a task with 1M input tokens, 90% of them cached, and 50K output tokens (900K cached, 50K uncached, 50K output):
| Model | Arithmetic | Credits |
|---|---|---|
| GPT-6 Luna | 0.9 × 0.25 + 0.05 × 2.5 + 0.05 × 12.5 | 0.98 |
| GPT-6.1 Terra | 0.9 × 1.25 + 0.05 × 25 + 0.05 × 125 | 8.63 |
| GPT-6.1 Sol | 0.9 × 2.5 + 0.05 × 50 + 0.05 × 250 | 17.25 |
| GPT-6 Sol | 0.9 × 5 + 0.05 × 50 + 0.05 × 250 | 19.50 |
| GPT-5.6 Sol | 0.9 × 10 + 0.05 × 100 + 0.05 × 500 | 39.00 |
| GPT-6 Astra | 0.9 × 25 + 0.05 × 250 + 0.05 × 1,250 | 97.50 |
Output dominates every row: on GPT-6.1 Sol, the 50K output tokens are 12.5 of the 17.25 credits. GPT-6.1 Sol uses less than half the credits of GPT-5.6 Sol for the same tokens, so moving off GPT-5.6 stretches the same plan allowance more than twice as far.
Plans and Speed Multipliers
| Plan | Price |
|---|---|
| Free | $0 a month; GPT-6 Luna at Standard speed in the desktop app, subject to rollout |
| Go | $8 a month |
| Plus | $20 a month |
| Pro | $100, $200 or $500 a month; Ultrafast only on Pro $500 |
| Business | $20 per user a month billed annually, or $25 billed monthly, for 2 or more users |
| Enterprise, Edu | Contact sales |
GPT-6.1 Sol is available in Codex on Plus and above (Enterprise and Edu admins must enable it). GPT-6.1 Terra is available on Plus, Pro, Business, Enterprise and Edu.
| Speed | Included plan usage | Purchased credits / Enterprise pay-as-you-go |
|---|---|---|
| Fast | 2.5x | 2x |
| Ultrafast (GPT-6 Astra and GPT-6.1 Sol) | 8x | 6x |
Fast mode burns included usage faster (2.5x) than it costs in purchased credits (2x). If you rely on Fast mode, the included allowance runs out sooner than a straight 2x estimate suggests.
Five-Hour Windows and Weekly Limits
OpenAI publishes estimated local messages per five-hour window for Plus and Standard Business:
| Model | Local messages per 5 hours |
|---|---|
| GPT-6 Astra | 5–45 |
| GPT-6.1 Sol | 15–160 |
| GPT-6.1 Terra | 35–285 |
| GPT-6 Sol | 15–150 |
| GPT-6 Luna | 350–3,000 |
The rules around these numbers:
- Pro currently has no five-hour limit.
- Weekly limits may also apply on top of the five-hour window. OpenAI does not publish the weekly figure; check Codex Settings > Usage for your remaining allowance.
- Local messages and cloud tasks share the plan’s allowance, and cloud tasks can use more of it.
- Enterprise and Edu with flexible pricing have no fixed rate limits. Without flexible pricing, those plans have the same per-seat limits as Plus for most features.
- Plus and Pro can buy credits after the included usage runs out, instead of waiting or upgrading. OpenAI does not publish one per-credit price; it depends on the plan or agreement.
So the accurate summary is: Pro has no five-hour cap, but no plan has an officially unlimited weekly allowance unless it is an Enterprise or Edu workspace on flexible pricing.
Subagents and Parallel Work
Codex lists subagents and custom agents as a feature. Each subagent is its own token-consuming workstream, so a task split across four subagents can use several times the credits of the same task run by one agent, even if it finishes sooner. Run cheap exploration subagents on GPT-6 Luna or GPT-6.1 Terra and keep the final synthesis on GPT-6.1 Sol.
GPT-5.5 Has Left Codex
GPT-5.5 was retired from Codex with ChatGPT sign-in on October 14, 2026 at noon PDT. API-key sessions were not affected, and GPT-5.5 is still on the API price list at $5 / $30.
When GPT-5.5 was OpenAI’s frontier model (April to July 2026), Codex used it for hard tasks, and its credit rate equalled GPT-5.6 Sol’s launch rate. The dedicated Codex API model was then, and still is, gpt-5.3-codex; there was never a gpt-5.5-codex model ID. GPT-5.3-Codex costs $1.75 input / $0.175 cached / $14 output per 1M tokens and has a 400K context window.
Codex With an API Key: What the Same Task Costs
With an API key, the 1M-token task above (900K cached, 50K uncached, 50K output) costs:
| Model | Arithmetic | API cost |
|---|---|---|
| GPT-6 Luna | 0.9 × $0.01 + 0.05 × $0.10 + 0.05 × $0.50 | $0.039 |
| DeepSeek V4.1 Flash, off-peak | 0.9 × $0.003 + 0.05 × $0.15 + 0.05 × $0.60 | $0.040 |
| DeepSeek V4.1 Flash, peak | 0.9 × $0.006 + 0.05 × $0.30 + 0.05 × $1.20 | $0.080 |
| GPT-6.1 Terra | 0.9 × $0.05 + 0.05 × $1 + 0.05 × $5 | $0.345 |
| GPT-6.1 Sol | 0.9 × $0.10 + 0.05 × $2 + 0.05 × $10 | $0.690 |
| GPT-5.3-Codex | 0.9 × $0.175 + 0.05 × $1.75 + 0.05 × $14 | $0.945 |
| GPT-5.6 Sol (promo) | 0.9 × $0.40 + 0.05 × $4 + 0.05 × $20 | $1.560 |
| GPT-5.5 | 0.9 × $0.50 + 0.05 × $5 + 0.05 × $30 | $2.200 |
API cache writes are left out for simplicity; on GPT-6 models they add 1.25x base input for each token written. GPT-6.1 Sol now costs less per task than the dedicated GPT-5.3-Codex model, so there is no price reason to pin gpt-5.3-codex unless a tool requires it. Codex CLI can also be pointed at other OpenAI-compatible providers; the DeepSeek V4 agent setup guide shows how.
Which Model Preserves Codex Usage Best?
| Work | Model | Credits per task above |
|---|---|---|
| Repository search, file summaries, classification, routine edits | GPT-6 Luna | 0.98 |
| Normal implementation, debugging, visual and computer-use tasks | GPT-6.1 Terra | 8.63 |
| Hard multi-file changes and long agent runs | GPT-6.1 Sol | 17.25 |
| The hardest tasks after Sol fails | GPT-6 Astra | 97.50 |
Running everything on Astra uses about 100 times the credits of Luna for the same tokens. A Luna → Terra → Sol escalation keeps most of the allowance for the tasks that need it.
For direct API pricing of these models, see the OpenAI API pricing overview, the GPT-6 pricing guide and the GPT-5.6 pricing guide. To compare Codex with other coding agents, read the AI coding agent cost comparison.
Sources
Checked November 16, 2026:
- Codex pricing and usage limits: credit rate card, plans, five-hour estimates, weekly limits, Fast and Ultrafast multipliers, GPT-5.5 retirement, API-key usage
- OpenAI API pricing: token prices for the API-key comparison
- DeepSeek API pricing: V4.1 Flash peak and off-peak rates
- July 2026 GPT-5.6 Codex credit rates are from our earlier version of this page.
Next step
Related Posts
AI Coding Agent Cost Comparison 2026: GPT-6, Opus 5.5, DeepSeek, and Kimi
2026-05-07
Codex + DeepSeek V4: How to Use DeepSeek in Codex CLI (Official Setup, 2026)
2026-11-16
How to Use GLM-5.3 in Claude Code (Official Z.AI Setup, 2026)
2026-11-16