/ palantir / blueprint
Updated 7/19/2026
A React-based UI toolkit for the web
Key Takeaways
- Claude Fable 5 leads with 55 of 59 tasks resolved and is the only model to solve one especially difficult task.
- GPT-5.6 Terra, GLM 5.2, Claude Opus 4.8, and Claude Opus 4.7 form the next performance tier at 52 of 59.
- Claude Fable 5 and GPT-5.6 Terra resolve 57 of 59 tasks together, while Terra is the strongest efficiency option at $0.49 per test, 526 seconds, and 0.60M tokens per test.
Cost / Test vs. Accuracy
Claude Fable 5 leads at 93.22% for $5.33 per test, while GPT-5.6 Terra reaches 88.14% for $0.49 per test.
Latency vs. Accuracy
GPT-5.6 Terra is the fastest model in the 52-of-59 tier at 526 seconds; Claude Fable 5 takes 754 seconds for three additional resolved tasks.
Average token use / test
Tasks with failures
| Model | 051361b | 08c0376 | 13866c7 | 147cbe3 | 24a4843 | 3c780bc | 42a0c1a | 4587c0a | 4c88a05 | 6273c37 | 632fbab | 8d59758 | 956f23b | b6310ce | b80e9b7 | b90ef52 | b9adf0a | c56b2a8 | c5e8bdc | c791bdc | cc283b0 | d51c90c | dc7c104 | e15084f | e7d418e | f46419e |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Claude Fable 5 | ||||||||||||||||||||||||||
| GPT-5.6 Terra | ||||||||||||||||||||||||||
| GLM 5.2 | ||||||||||||||||||||||||||
| Claude Opus 4.7 | ||||||||||||||||||||||||||
| Claude Opus 4.8 | ||||||||||||||||||||||||||
| GPT-5.6 Luna | ||||||||||||||||||||||||||
| Grok 4.5 | ||||||||||||||||||||||||||
| Claude Sonnet 5 | ||||||||||||||||||||||||||
| Kimi K3 | ||||||||||||||||||||||||||
| Gemini 3.5 Flash | ||||||||||||||||||||||||||
| GPT 5.5 | ||||||||||||||||||||||||||
| GPT-5.6 Sol | ||||||||||||||||||||||||||
| Muse Spark 1.1 | ||||||||||||||||||||||||||
| Gemini 3.1 Pro Preview (02/26) | ||||||||||||||||||||||||||
| Claude Haiku 4.5 (Nonthinking) |
Task detail
051361b08429Issue statement
DateRangeInput does not close its calendar popover when Escape is pressed
The DateRangeInput component (in @blueprintjs/datetime) opens a calendar popover when one of its two text inputs is focused. Standard overlay/popover behavior in the toolkit is that pressing the Escape key while interacting with the control dismisses the popover, but DateRangeInput ignores the Escape key entirely.
Steps to reproduce
- Render a
DateRangeInputand focus the start input so the calendar popover opens (isOpenbecomestrue). - Press the
Escapekey while the input is focused.
Expected behavior
Pressing Escape should close the popover (isOpen becomes false) and the input fields should no longer be focused.
Actual behavior
Nothing happens: the popover stays open and the input keeps focus.
The key-down handling for the range inputs should treat Escape as a request to dismiss the popover and blur the currently focused field.
Hidden tests
diff --git a/packages/datetime/test/components/dateRangeInputTests.tsx b/packages/datetime/test/components/dateRangeInputTests.tsxindex deb4c7210..018d4aeca 100644--- a/packages/datetime/test/components/dateRangeInputTests.tsx+++ b/packages/datetime/test/components/dateRangeInputTests.tsx@@ -602,6 +602,21 @@ describe("<DateRangeInput>", () => { expect(root.state("isOpen"), "popover closed at end").to.be.false; }); + it("pressing Escape closes the popover", () => {+ const { root } = wrap(<DateRangeInput {...DATE_FORMAT} value={[null, null]} />);+ root.setState({ isOpen: true });++ const startInput = getStartInput(root);+ startInput.simulate("focus");++ expect(root.state("isOpen")).to.be.true;++ startInput.simulate("keydown", { key: "Escape" });++ expect(root.state("isOpen")).to.be.false;+ expect(isStartInputFocused(root)).to.be.false;+ });+ it("Clicking a date invokes onChange with the new date range and updates the input fields", () => { const defaultValue = [START_DATE, null] as DateRange;
Head-to-head
Accuracy
93.22%Δ 5.08%
88.14%
Cost / test
$5.33
$0.49Δ $4.84
Latency
754s
526sΔ 229s
Cost distribution
Latency distribution
Input token distribution
Output token distribution
Task outcomes
59 tasks