SShortSingh.
TopTechnologyAI & MLStartupsProgrammingScienceBusinessFinanceCrypto & Web3World
Back to feed
Programming

Benchmarking Startup Latency of Coding Agent CLIs with hyperfine

0
·7 views

A DEV Community guide explains that coding agent CLIs can take noticeable time before a prompt reaches the model, especially when MCP servers are slow to respond. It notes documented wait limits: Claude Code waits up to 30 seconds for MCP servers by default, Codex waits 1,000 milliseconds for optional servers, and OpenCode allows 5,000 milliseconds for tool fetching. The author proposes using hyperfine, a benchmarking tool, to measure startup overhead layer by layer across Claude Code, Codex, GitHub Copilot CLI, and OpenCode in headless mode. The piece stresses that real timings depend on each user's own setup, so the documented figures are only ceilings. It also covers installation steps and notes that hyperfine 2.0 runs commands without a shell by default.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Log in to join the discussion and vote.

Log in

Related stories

0
ProgrammingDEV Community ·

HustleRadar Uses Live Resale Evidence to Check Whether a Discount Is a Real Flip

The developer behind HustleRadar argues that a crossed-out reference price alone does not show a profitable resale opportunity. The open-source tool asks buyers to enter a product, their actual purchase price, and an optional reference price. It then gathers shopping and retailer evidence, normalizes product variants, and calculates market prices and fee-adjusted resale margins. Gemma 2 explains the results, with a deterministic fallback if the model is unavailable. The project is aimed at electronics flippers and small resellers, and its demo report is labelled as a reproducible scenario rather than live pricing.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Grid-based caching for map radius searches, and the edge cases at the poles and antimeridian

A business-campaign tool needs to load local places into its database for each customer's search area, and reloading for every request would be wasteful. The solution treats half-degree grid cells, rather than the circle itself, as the unit of work, so overlapping requests share cached data under common keys. Latitude conversion is straightforward, but longitude distance shrinks toward the poles, so the code scales by the cosine of latitude and caps the stretch to avoid division blow-ups near the pole. Whole countries are stored as a single area with a shared key and a bounding box, with a country-code filter to exclude places that fall outside the border. Missing cells are then grouped into two-degree blocks so that imports run in fewer batches.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Team reverses its adaptive daily API cap after a week, moving limits to per-plan and per-campaign

A developer team shipped a daily API cap that scaled with the previous week's busiest day, then removed it seven days later. The mechanism used one shared counter for the whole product, so a spike in one account could block unrelated paying customers from finding creators for the rest of the day. The team now sets limits per plan, listed on the pricing page, and per campaign, so the account that hits a limit is the one that caused it. Provider-imposed quotas, such as Google's, remain shared because the API key is shared.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Solo Developer Shares Lessons From Building a Maze Game Alone

A developer who left a midsize studio describes building a solo maze game and finds that staying organized across every role was harder than the coding itself. They narrowed the project to a minimal feature set centered on a single gameplay loop, covering maze generation, movement, artifact pickup, and basic ghost AI. After weighing Unity, Godot, and a custom engine, they chose Godot for its fast editor, Python-like scripting, and built-in export tools. Frequent Git commits, automated build scripts, unit tests for the maze generator, and two-week sprints kept the work structured. A friend's playtest feedback then prompted quick fixes to visual cues and respawn invincibility.

0 comments Read more at DEV Community