While tuning Codex CLI execution speed, you will notice it has a /fast command. Claude Code also has /fast. The name is the same, but the mechanism and cost are different.
Codex: Spend Quota for Speed
Codex fast mode adds a field to the API request that tells the OpenAI backend: “handle this request through the priority path.” Officially, speed improves by about 1.5x, with lower and more stable latency. The model does not change, and reasoning depth is not reduced.
The cost is 2x quota consumption. Whether you use ChatGPT subscription credits or included usage on the Pro plan, turning on fast burns quota at double speed.
In practice, Codex quota is already generous. Especially during the promo period where the Pro $100 plan has doubled quota, the 2x fast-mode consumption is partly offset, so it does not feel too painful.
Important: fast mode is not enabled by default. You must manually run /fast on. Codex CLI’s feature flag exposes the /fast command by default, but that only means the command is available; it does not mean fast mode is running.
Codex also has something called GPT-5.3-Codex-Spark, which is a completely different “fast but weaker” model. It is not the same as fast mode.
Claude Code: Same-Model Acceleration, but 6x Price
Claude Code’s /fast takes a different path. It uses another API configuration to make output about 2.5x faster, and the official wording says quality and capability stay the same.
The cost is on the pricing table: input jumps from $5 to $30, output from $25 to $150 per 1M tokens. That is exactly 6x. For subscription users, fast mode also counts as extra usage, not included plan quota.
One easy-to-miss prerequisite: when /fast is enabled and you are not already using Opus 4.6, it automatically switches to Opus 4.6. Turning fast off does not automatically switch back to the previous model. So “same model with another configuration” is fully true only if you were already on Opus 4.6.
If you enable fast mode mid-conversation, the already-cached context in that session is also billed at fast pricing. If you are going to use it, enabling it at the start of the session is cheaper.
Same Word, Different Tradeoff
| Codex /fast | Claude Code /fast | |
|---|---|---|
| Mechanism | Priority processing + acceleration | High-speed API configuration |
| Speed gain | About 1.5x | About 2.5x |
| Model | Unchanged | Unchanged, but may auto-switch to Opus 4.6 |
| Quality | Unchanged | Unchanged |
| Extra cost | 2x quota consumption | 6x token price, counted as extra usage |
| Default | Off; manual enable required | Off |
Both are off by default. Codex fast trades quota for speed, and if quota is abundant, enabling it can be worth it. Claude Code fast is real money at 6x price; every use moves the bill.
I had to read Codex’s Rust source and compare official docs to align the real behavior of both tools. The same feature name can mean very different things in different products. Next time a name looks familiar, do not assume. Check the implementation.
Further Reading
- Complete Claude Code Tutorial
- Claude Code vs Cursor
- Claude Opus vs Sonnet
- Claude Code + Codex Collaboration Playbook
Penchan’s Take
Claude Code and Codex are both daily drivers for me, and I have tried /fast on both. Codex fast feels more practical. The quota is there, and the difference is noticeable. On Claude Code, seeing the 6x price made me back out. Normal Opus / Sonnet speed is already enough for writing articles and debugging, so paying that kind of token bill for a bit more speed rarely makes sense. Decide whether to use fast at the start of the session; switching midstream is where billing surprises happen.
FAQ
Q: What does Codex fast mode do?
Codex fast mode adds a field to the API request telling the OpenAI backend to use the priority path. The model does not change, reasoning depth does not drop, and quality does not change. It is essentially queue priority.
Q: What does Claude Code fast mode do?
Claude Code fast mode uses a different API configuration to make the same model output about 2.5x faster, but token price is 6x higher: input $5→$30 and output $25→$150 per 1M tokens.
Q: Is Codex fast mode on by default?
No. Codex CLI’s feature flag exposes the /fast command by default, but actual fast mode is not enabled automatically. You need to run /fast on manually to use the priority path.
— Penchan