LoginSign Up
← Back to all sparks
A

Aider

AI-ASSISTANTS
Velocity0.0

AI pair programming in your terminal

aider.chat ↗

Aider's changelog reads as a model-benchmark ledger, with the CLI a quiet beneficiary.

ai-codingbenchmarksarchitect-editormodel-routingcli-toolingopen-source
◆Current state
Aider is a terminal-based AI pair programmer whose public cadence is dominated by posts on its own polyglot leaderboard rather than feature releases. The recent stream is almost entirely model evaluations — Qwen3, Gemini 2.5 Pro, R1+Sonnet — plus errata and provider-availability advisories. Genuine product changes, like the uv-based installer and the polyglot benchmark itself, surface only intermittently between leaderboard updates.
◆Where it's heading
Aider is consolidating its position as a neutral scoreboard for coding LLMs, with the architect/editor split — a reasoning model paired with an editing model — as its core technical bet. The benchmark-post cadence will keep tracking each major model launch, while real product work on installation and model routing ships quietly underneath. The signal-to-release ratio is low: most entries inform rather than change the tool.
◆Prediction
The next entries are most likely benchmark results for whatever frontier model ships next, with occasional install or provider-routing fixes in between.

◆Recent moves

  1. 1y ago

    Qwen3 benchmark results

    Another leaderboard update — Qwen3 run through aider's polyglot benchmark. It extends the model-evaluation cadence that dominates aider's stream but changes nothing about the tool itself.

    View source ↗
  2. 1y ago

    Gemini 2.5 Pro Preview 03-25 benchmark cost

    A one-line correction to a previously reported Gemini 2.5 Pro benchmark cost. Pure errata on the leaderboard, with no product impact.

    View source ↗
  3. 1y ago

    Alternative DeepSeek V3 providers

    Advisory content pointing users to alternative DeepSeek V3 providers during an API outage. Useful context, but a routing tip rather than a change to aider.

    View source ↗
  4. 1y ago

    R1+Sonnet set SOTA on aider’s polyglot benchmark

    R1 as architect paired with Sonnet as editor tops the polyglot benchmark at a fraction of o1's cost — concrete validation of aider's architect/editor split as its central design bet. Still a benchmark post, but one that reinforces the product's core thesis.

    View source ↗
  5. 1y ago

    Using uv as an installer

    One of the few genuine product changes in the stream: aider moves its install path onto uv, bundling dependencies and Python 3.12 into an isolated environment. It lowers the setup friction that dogs Python CLI distribution.

    View source ↗
  6. 1y ago

    o1 tops aider’s new polyglot leaderboard

    Aider introduces its harder, multi-language polyglot benchmark with o1 taking the top slot. The new leaderboard becomes the backdrop for nearly every subsequent entry — infrastructure that underpins the product's benchmark-authority positioning.

    View source ↗