GPT-5.6 Sol vs GPT-5.3 Codex
Green marks the cheaper side per row — informational only, never a verdict for your workload
| API status | Available | Deprecated — shuts down 2027-04-01 |
|---|---|---|
| Context window | 1,050,000 tokens | 400,000 tokens |
| Knowledge cutoff | Feb 16, 2026 | Aug 31, 2025 |
| Standard input | $4.00/MTok | $1.75/MTok |
| Standard cached input | $0.40/MTok | $0.175/MTok |
| Standard output | $20.00/MTok | $14.00/MTok |
| Cache writes | Write rate $5.00/MTok | No surcharge |
| Reasoning effort levels | none, low, medium (default), high, xhigh, max | low, medium, high, xhigh |
| Faster tiers offered | Fast | Fast |
| Long-context repricing rule | Published | Not published |
| Build tier rate limit | 5,000 RPM · 1,000,000 TPM | 5,000 RPM · 1,000,000 TPM |
| ChatGPT credits per 1M (in / out) | 100 / 500 | No credit rate |
| Codex with ChatGPT sign-in | No retirement announced | Deprecated |
GPT-5.6 Sol: promotional pricing — OpenAI lowered GPT-5.6 Sol's prices on 2026-08-21 and says the promotional pricing is available at least through 2026-11-21. In ChatGPT plans it applies to usage paid with purchased credits; included plan usage and the 5-hour and weekly limits did not change.
This was the site's founding comparison: the Codex CLI's default model against the model named for Codex. Both halves of that framing have moved. The CLI now defaults to GPT-6.1 Sol, and GPT-5.3 Codex is deprecated — in Codex for ChatGPT sign-in and, since October 1, 2026, on the API, which shuts it down on April 1, 2027. What remains is a useful API-key comparison, and the pattern it taught still applies to the new default.
The axes that differ
Context window: Sol's is much larger. Knowledge cutoff: Sol's is several months newer. Price: GPT-5.3 Codex is cheaper on input, cached input and output, even against Sol's promotional rates. Larger window and newer knowledge cost more per token here, as they usually do.
What doesn't differ
Both sit in the same rate-limit table, tier for tier, so choosing one over the other costs nothing in throughput.
Asymmetries worth knowing
Sol publishes the long-context rule, Batch and Flex prices, and a cache-write rate; GPT-5.3 Codex publishes none of those — Standard and Fast only, with the long-context rule recorded as not published. Both publish reasoning-effort levels. This page once said only the Codex model did; that was a parsing failure on this site's side, not OpenAI's silence. Sol accepts none and max; GPT-5.3 Codex runs low to xhigh.
Caching narrows the gap further
GPT-5.3 Codex's cached-input rate is lower than GPT-5.6 Sol's, and it has no cache-write rate at all: written tokens bill at its ordinary input rate, while Sol bills writes at 1.25x its input rate. A workload that writes many short-lived prefixes pays proportionally more on Sol than its headline input price suggests.
The promotional asterisk
Sol's current prices are promotional, available at least through November 21, 2026, and OpenAI hasn't said what follows. Every gap on this page was measured against those rates; if Sol's prices rise afterwards, GPT-5.3 Codex's advantage grows with them.
How to decide
On an API key, choose Sol for large accumulated contexts and recent knowledge, GPT-5.3 Codex only when your sessions fit its window, cost dominates, and the work will be off it before its API shutdown. In Codex with ChatGPT sign-in, neither is the current default; compare against GPT-6.1 Sol instead.
Verified 2026-10-28 against CodexHow facts module (src/data/facts/) — see /about/#accuracy.