7 Local LLM Families To Replace Claude/Codex (for everyday tasks)

There are 7 open-source model families you can run locally that are now delivering real-world performance surprisingly close to commercial-grade models on many practical tasks.

If you don’t have a Medium membership, you can read the full article for free.

Besides leaderboard headlines, I will walk through local setups (e.g., Claude Code and Codex), a hardware tier guide (NVIDIA GPUs and Mac setups), and the operational gotchas, failure modes and trade-offs: latency, memory footprint, tool use, coding reliability, and instruction-following quality.

There are two big reasons why this is happening.

First, open-source models can now handle a meaningful share of everyday work. For a lot of coding, writing, automation, and agentic tasks, you no longer need to default to Opus 4.6 for every step.

Second, a local fallback helps preserve expensive credits. If you are running agent swarms, iterating heavily, or delegating lots of sub-tasks, local models are one of the easiest ways to avoid burning through your Opus 4.6 or Codex 5.3 credits on work that does not actually require them.

Moreover, if you had told me a year ago that an open-source model would land within striking distance of top closed models on SWE-bench Verified, I would have been politely skeptical.

But that performance gap has narrowed dramatically for a set of well-scoped tasks.