By Claude Code v2.1.142 shipped on May 14 and quietly changed the default model for the /fast command from Opus 4.6 to Opus 4.7. No banner. No in-app notification. Just a changelog entry most developers haven’t read. The practical consequence: if you’ve been using /fast in Claude Code this week, you’ve been running on Opus 4.7 at rates that are 6x higher than standard — and the math on whether that’s worth it is more complicated than the headline speed gain suggests.
What Fast Mode Actually Is
Fast Mode is a research preview feature (beta) that delivers the same Claude Opus model with a faster inference configuration. On Opus 4.7, that means up to 2.5x more output tokens per second — same intelligence, same benchmarks, same 1 million-token context window. Anthropic first launched it for Opus 4.6 in February; the extension to Opus 4.7 arrived May 13, followed by the Claude Code default switch a day later.
The performance numbers hold up: Opus 4.7 scores 87.6% on SWE-bench Verified and leads MCP-Atlas at 77.3%, ahead of GPT-5.4 (68.1%) and Gemini 3.1 Pro (73.9%). Fast Mode doesn’t change any of that — the brain is identical. What you’re buying is faster delivery of the same output.
The Real Cost Calculation
The pricing is steep:
- Standard Opus 4.7: $5/M input tokens, $25/M output tokens
- Fast Opus 4.7: $30/M input tokens, $150/M output tokens
That 6x multiplier is the headline, but there’s a compounding factor most coverage misses. Opus 4.7 ships with a new tokenizer that uses up to 35% more tokens on the same input compared to earlier Claude models. So even at “unchanged” standard list prices, you’re already paying more per effective unit of work. Stack the 6x fast mode premium on top of a tokenizer that inflates token counts, and the real cost delta is larger than the 6x figure implies.
To make it concrete: a developer running 10 interactive sessions per day at roughly 50,000 tokens each pays about $12.50/day in standard mode. The same workload in fast mode runs $75/day — a difference of $62.50 daily, or roughly $1,900/month, purely for faster token delivery.
The Cache Trap Nobody Talks About
There’s a subtle cost risk in Fast Mode that Anthropic doesn’t lead with: switching fast mode on mid-conversation reprices the entire conversation context at fast mode rates, not just from the switching point forward. Combined with prompt cache invalidation that can cause up to a 5x cost spike for that turn, toggling fast mode mid-session is an expensive mistake.
The practical rule is simple: decide at session start. Commit to fast or standard and don’t switch. On Team and Enterprise plans, fast mode is disabled by default and requires an admin toggle — which forces that decision deliberately instead of letting premium costs accumulate quietly.
Where Fast Mode Makes Sense
The use case for Fast Mode is genuinely narrow but real. Interactive debugging sessions, rapid design iteration, real-time UI generation, and any workflow where a developer is actively watching responses and iterating immediately — these are the scenarios where 2.5x speed compounds across dozens of micro-interactions per session and translates into actual time savings.
A senior developer billing at $200/hour who recovers even 30 minutes of waiting time daily can justify the premium on pure economics. The problem is that most workflows don’t fit this profile.
For batch jobs, CI/CD pipelines, long autonomous agents, or any task where you walk away while Claude runs — fast mode is cost waste. You’re paying for a speed you don’t experience. Standard Opus 4.7 delivers the same output at 1/6 the price with no degradation in quality.
How to Enable or Opt Out
To use Fast Mode via the API, set speed: "fast" with model claude-opus-4-7 and add the beta header anthropic-beta: fast-mode-2026-02-01. The response includes usage.speed confirming which mode ran.
import anthropic
client = anthropic.Anthropic()
message = client.beta.messages.create(
model="claude-opus-4-7",
max_tokens=1024,
speed="fast",
betas=["fast-mode-2026-02-01"],
messages=[{"role": "user", "content": "Debug this code..."}]
)
# Verify which mode ran
print(message.usage.speed) # "fast" or "standard"Fast Mode is available on the direct Anthropic API, Claude Code, Cursor, Windsurf, v0, Warp, and Vercel AI Gateway — but not on the Batch API, Amazon Bedrock, Google Vertex AI, or Azure Foundry. If you want to stick with Opus 4.6 for your /fast sessions in Claude Code, set the environment variable CLAUDE_CODE_OPUS_4_6_FAST_MODE_OVERRIDE=1.
The Verdict
Fast Mode for Opus 4.7 is a well-executed feature for a specific type of developer: someone who spends most of the day in interactive back-and-forth with Claude and whose time cost genuinely exceeds the API premium. For everyone else — the majority of API users and most Claude Code workflows — the 6x price jump buys speed that doesn’t translate into real productivity gains. The May 14 default change deserved a more prominent heads-up from Anthropic. Developers shouldn’t have to dig through a GitHub release to find out their billing profile changed.
