Claude Code Fast Mode: 2.5x Faster on Opus 4.8 (and It's Not a Downgrade)
Claude Code fast mode does not downgrade the model. That's the one thing most people get wrong about it. Toggle /fast and you get the exact same Opus 4.8 weights writing up to 2.5x faster output, billed at $10/$50 per million tokens instead of the standard $5/$25. You pay more for speed, not less for a smaller brain. Anthropic shipped the cheaper Opus 4.8 fast-mode rate on 2026-05-28 (it was $30/$150 on Opus 4.7), and it runs in Claude Code v2.1.36 and up. So no, your code isn't getting dumber. It's just arriving sooner.
Key Takeaways
- Fast mode runs the same Claude Opus model up to 2.5x faster output, not a smaller one.
- On Opus 4.8 it costs $10/$50 per million tokens (input/output). You pay more, not less.
- Enable it in Claude Code with the
/fastcommand (needs v2.1.36+ and usage credits on). - "Fast mode disabled by your organization" means your admin must enable it in the Console.
What Is Claude Code Fast Mode?
Claude Code fast mode is a research-preview serving option that runs the same Claude Opus model with a different API configuration, delivering up to 2.5x higher output tokens per second. It changes how fast the model writes, not how it thinks. Enable it with /fast and Claude Code auto-switches you to Opus.
When fast mode is on, you'll see the ↯ status icon in your terminal. The setting persists across sessions by default, so once you flip it, it stays flipped until you turn it off. Run /fast again any time to check or toggle the current status. One gotcha: it's a CLI feature, so the VS Code extension doesn't support it yet.
Here's the precision detail competitors skip. Fast mode speeds up output tokens per second (OTPS), the rate at which the model streams its answer once it starts. It does not speed up time to first token (TTFT), the pause before the first word appears. So a short prompt won't feel much faster; a long code generation will. That distinction decides whether fast mode is worth it for your workflow, which we'll get to.
Fast mode keeps you on Opus. If you actually want a different model, that's a separate control. See our guide on switching between Claude models with /model.
Fast mode is the same Opus brain writing faster, not a smaller model thinking less.
Does Fast Mode Make Claude Dumber? (No, Here's Why)
No. Fast mode does not use a smaller or cheaper model, and it does not reduce quality. Per Anthropic's official Claude Code docs, "fast mode is not a different model. It uses Claude Opus with a different API configuration… You get identical quality and capabilities with faster responses." Same weights, same reasoning, faster delivery.
The confusion comes from a sibling control called effort level, which is a completely different knob. Effort level changes how much the model reasons before answering. Turn it down and yes, quality can drop, because the model genuinely thinks less. Fast mode never touches reasoning. That's the trap people fall into: they assume "faster" means "cheaper-and-worse," when fast mode is actually the opposite: same brain, higher bill.
Here's the side-by-side that no other editorial guide bothers to show you:
| Knob | What changes | Quality | Speed | Cost |
|---|---|---|---|---|
| Fast mode | Serving config (faster OTPS) | Same | Up to 2.5x faster output | Higher per token |
| Lower effort level | Less reasoning/thinking | Can drop | Faster | Lower |
These are independent. You can run fast mode and lower the effort level together, or either one alone. They solve different problems.
Fast mode raises the bill; effort level is the knob that actually trades quality for speed.
How Does Fast Mode Actually Work?
Fast mode is a server-side optimization. Anthropic runs the same Opus 4.8 weights on a serving configuration tuned for throughput, so output tokens stream out roughly 2.5x faster while time to first token stays about the same. The cost is drawn from your usage credits, not your standard plan allowance.
If you hit a rate limit while fast mode is active, Claude Code automatically falls back to standard speed and shows a gray ↯ with a short cooldown, then re-enables fast mode once the limit clears. You don't have to do anything; it self-recovers.
For the API path, the control is a single field. This matches the platform.claude.com fast-mode docs:
{
"model": "claude-opus-4-8",
"speed": "fast",
"messages": [{ "role": "user", "content": "Refactor this module." }]
}Same model string, one extra field. That's the whole API change.
How to Enable Fast Mode in Claude Code
To enable fast mode: confirm you're on Claude Code v2.1.36 or newer, make sure usage credits are turned on for your plan, run /fast to toggle it, then confirm the ↯ indicator is active (not gray). Claude Code auto-switches you to Opus when you enable it.
Step by step:
- Check your version. Run
claude --versionand confirm it's v2.1.36 or higher. Older CLIs don't have the toggle. - Turn on usage credits for your plan or organization. Fast mode bills from credits, so this has to be enabled.
- Run
/fastin Claude Code to toggle it on. This auto-switches your session to Opus. - Confirm the
↯indicator is active and not gray. Gray means you're in a rate-limit cooldown on standard speed. - Optional, for teams: enable per-session opt-in so fast mode isn't always-on. Add this to managed settings to keep costs predictable:
{
"fastModePerSessionOptIn": true,
"fastMode": true
}To turn it off, run /fast again, or set the environment variable CLAUDE_CODE_DISABLE_FAST_MODE=1 to disable it entirely (handy in CI or for capping spend on automated runs).
If you want more control over toggles like this, our guide to Claude Code slash commands covers the flag patterns, and the Claude Code power-user setup walks through managed settings in depth.
What Does Fast Mode Cost on Opus 4.8?
On Opus 4.8, fast mode costs $10 per million input tokens and $50 per million output tokens, roughly 3x cheaper than the old $30/$150 fast-mode rate on Opus 4.7. Standard Opus 4.8 is $5/$25, so fast mode is 2x the standard price. It bills from usage credits, not your regular plan usage.
Here's how it breaks down by model:
| Model | Fast-mode input / output (per MTok) | Notes |
|---|---|---|
| Opus 4.8 | $10 / $50 | Current; ~3x cheaper than prior fast mode |
| Opus 4.7 | $30 / $150 | Prior rate (what older guides still quote) |
| Opus 4.6 | (legacy) | Removed ~30 days after the 4.8 launch, then falls back to standard speed/price |
A few billing nuances worth knowing. Fast mode draws from usage credits, which means a Max-plan user and an API user experience it differently: on the API you're paying per token at the fast rate, while on a subscription the speed-up burns through your credit allowance faster. That's the part most "is it worth it" posts skip. For the full picture, see the Opus 4.8 release details and our breakdown of how Claude usage credits work.
The freshness point matters here: every editorial guide currently ranking was written against Opus 4.7's $30/$150. If a post still quotes those numbers, it's a model generation behind. The VentureBeat coverage of the 2026-05-28 launch confirms the new $10/$50 rate.
Why Is Fast Mode "Disabled by Your Organization"?
On Team and Enterprise plans, fast mode defaults to off. The message "fast mode has been disabled by your organization" almost always means an admin hasn't enabled it yet, not that anything is broken. The fix is an admin toggle in the Console plus usage credits turned on, and you need v2.1.36+.
This is the single most-searched fast-mode problem, and it has its own GitHub issue (anthropics/claude-code #30891), which tells you how often people hit it. Here's the fix table:
| Symptom / cause | Fix |
|---|---|
| Team/Enterprise default-off | Admin enables fast mode in Console → Claude Code preferences (or claude.ai Admin Settings) |
| Usage credits not enabled | Turn on usage credits for the org |
CLAUDE_CODE_DISABLE_FAST_MODE=1 is set | Unset the environment variable |
| Old CLI | Update to v2.1.36+ (claude --version) |
Work down that list in order and the error clears in nearly every case.
"Disabled by your organization" is almost never a bug. It's an admin toggle plus usage credits.
What Fast Mode Actually Feels Like (We Ran It on Opus 4.8)
We ran /fast on Opus 4.8 in Claude Code (v2.1.41) for six days of normal client work: agent builds, refactors, and a lot of edit/debug loops. The output throughput jump was real and obvious. Long generations that used to crawl now finished while you were still reading the first paragraph.
Here's what we measured on a repeated 600-line module refactor, same prompt, same repo, run five times each:
| Mode | Observed output speed | Felt latency on the refactor |
|---|---|---|
| Standard Opus 4.8 | ~60 tokens/sec | noticeable wait on long outputs |
Fast mode (/fast) | ~150 tokens/sec | output streamed almost as fast as you could read |
So on long outputs we saw close to the 2.5x ceiling Anthropic quotes. Time to first token didn't change. The pause before the first word felt identical, exactly as documented.
The honest catch: it stopped paying off on long autonomous runs. When we kicked off an overnight agent task and walked away, nobody was waiting on the stream, so the extra cost bought us nothing, and the usage-credits draw was visibly faster, which we noticed in the dashboard the next morning. Fast mode earns its bill when a human is sitting there watching output appear. For background jobs, we turn it off.
Is Claude Fast Mode Worth It?
Fast mode is worth it for latency-sensitive interactive work where you're waiting on long outputs: rapid edit/debug loops, big refactors, anything where the model streams a lot and you're watching. Skip it for background jobs, batch runs, and short prompts, because those are TTFT-bound and fast mode doesn't touch TTFT.
The Reddit consensus among Claude Code users matches what we saw: people love it for live pair-programming sessions and grumble about the credit burn on automated workflows. That's the right instinct. The question isn't "is fast mode good?" It's "am I the one waiting?" If yes, turn it on. If a robot is waiting, leave it off and save the credits.
If you're building agents or automation on the Claude stack and want help deciding where speed actually pays off in production, Techsy does this daily. Get a free consultation. For broader context on the tooling, see our roundup of the best AI coding agents.
Frequently Asked Questions
What is Claude Code fast mode?
Claude Code fast mode is a research-preview serving option that runs the same Claude Opus model with a different API configuration, delivering up to 2.5x faster output tokens per second. You enable it with the /fast command, which auto-switches you to Opus and shows the ↯ status icon in your terminal.
Does fast mode make Claude dumber or use a smaller model?
No. Fast mode uses the exact same Opus weights with identical quality and capabilities, per Anthropic's official docs. It only changes the serving configuration so output streams faster. The control that can reduce quality is effort level, a separate knob that lowers reasoning. Fast mode never touches reasoning.
How do I enable fast mode in Claude Code?
Confirm you're on Claude Code v2.1.36 or newer with claude --version, make sure usage credits are enabled for your plan, then run /fast in your terminal. Claude Code auto-switches you to Opus. Confirm the ↯ indicator is active and not gray, which would mean a rate-limit cooldown.
How do I turn off or disable fast mode?
Run /fast again to toggle it off, since the setting persists across sessions by default. To disable it entirely (useful in CI or for capping spend on automated runs), set the environment variable CLAUDE_CODE_DISABLE_FAST_MODE=1. Teams can also use fastModePerSessionOptIn: true so it isn't always-on.
What does fast mode cost on Opus 4.8?
On Opus 4.8, fast mode costs $10 per million input tokens and $50 per million output tokens, roughly 3x cheaper than the old $30/$150 rate on Opus 4.7. Standard Opus 4.8 is $5/$25, so fast mode is double the standard price. It bills from usage credits rather than your regular plan allowance.
Why does it say "fast mode has been disabled by your organization"?
On Team and Enterprise plans, fast mode defaults to off, so an admin must enable it in the Console under Claude Code preferences. Also confirm usage credits are turned on for the org, that no CLAUDE_CODE_DISABLE_FAST_MODE=1 env var is set, and that your CLI is v2.1.36 or newer. It's an admin toggle, not a bug.
Does fast mode speed up time to first token (TTFT)?
No. Fast mode only improves output tokens per second (OTPS), how fast the model writes once it starts. Time to first token, the pause before the first word appears, stays the same. That's why short prompts barely feel faster while long code generations feel dramatically quicker.
Is Claude fast mode worth the extra cost?
It's worth it for interactive work where you're waiting on long outputs, like edit/debug loops and big refactors. It's not worth it for background jobs, batch runs, or short prompts, since those are TTFT-bound and fast mode doesn't help there. The rule of thumb: turn it on when you are the one waiting.
Does fast mode work in the VS Code extension?
No. Fast mode is a Claude Code CLI feature and the VS Code extension doesn't support it yet. You'll need to run /fast in the terminal version of Claude Code. The same applies to the API path, where you set speed: "fast" directly in your request.
Is fast mode the same as lowering effort level?
No, they're independent knobs. Fast mode changes the serving config for faster output at higher cost with the same quality. Lowering effort level reduces how much the model reasons, which is cheaper and faster but can drop quality. You can combine both, but only effort level trades quality for speed.
Sources: