Your Claude budget_tokens Code Is Broken on Opus 5 — Migrate to adaptive + effort in 3 Lines

Chris Harper

2 min read

Jul 31, 2026 · 12:13 UTC

Workflow

Claude Code

Best Practices

Claude Opus 5 (launched July 24) rejects thinking: {type: "enabled", budget_tokens: N} with a 400 error — three lines migrate any existing code to the new adaptive mode with output_config.effort.

If you're calling Claude's thinking API today with budget_tokens, your code breaks the moment you switch to Opus 5 (or any 4.7+ model). Here's the exact migration.

What broke

# 400 error on Opus 5 and all models >= 4.7:
client.messages.create(
    model="claude-opus-5",
    max_tokens=16000,
    thinking={"type": "enabled", "budget_tokens": 10000},  # rejected
    messages=[...]
)

The fix

# Correct pattern for Opus 5 and all 4.7+ models:
client.messages.create(
    model="claude-opus-5",
    max_tokens=16000,
    thinking={"type": "adaptive"},      # or omit entirely on Opus 5
    output_config={"effort": "high"},   # low | medium | high | xhigh | max
    messages=[...]
)

On Claude Opus 5, Sonnet 5, Fable 5, and Mythos 5, thinking is on by default — you can omit the thinking field entirely if high effort works for your use case.

Migration table

Model	Thinking default	`budget_tokens`?	`effort`?
Opus 5, Sonnet 5, Fable 5, Mythos 5	On by default	400 error	Yes
Opus 4.8, 4.7	Off	400 error	Yes (needed to enable)
Sonnet 4.6, Opus 4.6	Off	Deprecated but accepted	Yes (preferred)
Haiku 4.5, Sonnet 4.5	Off	Required (`type: "enabled"` only)	No
Opus 4.5	Off	Required	Yes

Turn thinking off on Opus 5 (for fast, cheap responses)

client.messages.create(
    model="claude-opus-5",
    max_tokens=4096,
    thinking={"type": "disabled"},
    output_config={"effort": "low"},   # "disabled" + "xhigh"/"max" = 400 error
    messages=[...]
)

Cannot combine disabled with xhigh or max effort — that returns 400. On Fable 5 and Mythos 5, disabled is rejected entirely.

Streaming (required when max_tokens > 21,333 on Opus 5)

with client.messages.stream(
    model="claude-opus-5",
    max_tokens=64000,
    output_config={"effort": "max"},
    messages=[...]
) as stream:
    for text in stream.text_stream:
        print(text, end="", flush=True)
    response = stream.get_final_message()

To see thinking tokens alongside the response, add thinking={"type": "adaptive", "display": "summarized"} — thinking streams as thinking_delta events before text starts.