We ranked every major LLM by BenchLM's current coding formula — SWE-Rebench, SWE-bench Pro, LiveCodeBench, and SWE-bench Verified. Here's which models actually come out on top and why.
Share This Report
Copy the link, post it, or save a PDF version.
GPT-5.4 Pro is currently the top-ranked LLM for coding on BenchLM at 88.3, with Claude Opus 4.6 at 79.3 and Gemini 3.1 Pro at 77.8 close behind. The important change is methodological: BenchLM now gives real weight to SWE-Rebench in addition to SWE-bench Pro, LiveCodeBench, and SWE-bench Verified.
That change matters because it downweights one-off spikes and rewards fresher repository-style engineering signals. Models that looked artificially dominant when SWE-Rebench was ignored no longer sit at the top by default.
BenchLM's current coding score weights:
Here are the current top coding rows.
| Rank | Model | Type | SWE-Rebench | SWE-bench Pro | LiveCodeBench | Coding score |
|---|---|---|---|---|---|---|
| 1 | GPT-5.4 Pro | Reasoning | — | 89 | 86 | 88.3 |
| 2 | Claude Opus 4.6 | Non-Reasoning | 65.3 | — | 76 | 79.3 |
| 3 | Gemini 3.1 Pro | Non-Reasoning | 62.3 | 72 | 71 | 77.8 |
| 4 | GPT-5.4 | Reasoning | — | 57.7 | 84 | 76.1 |
| 5 | GPT-5.2 | Reasoning | — | 55.6 | 79 | 75.6 |
| 6 | GPT-5.3 Codex | Reasoning | 58.2 | 56.8 | 85 | 75.1 |
| 7 | GPT-5.1-Codex-Max | Reasoning | — | 84 | 67 | 74.2 |
| 8 | Claude Sonnet 4.6 | Non-Reasoning | 60.7 | — | — | 74.2 |
| 9 | Grok 4.1 | Non-Reasoning | — | — | 73 | 73.9 |
| 10 | GPT-5.2-Codex | Reasoning | 56.8 | 86 | 66 | 73.2 |
Full rankings with filters: Best LLMs for Coding.
Look at the HumanEval column. Six models score 91. Two more score 94-95. The benchmark has a ceiling problem — it tests function-level Python generation, and frontier models have gotten too good at it. HumanEval now tells you almost nothing about whether Model A is better than Model B at real coding work.
SWE-bench Verified and LiveCodeBench are where the actual separation happens. SWE-bench tests multi-file bug fixes in real GitHub repos. LiveCodeBench pulls fresh competitive programming problems so models can't memorize them. The spread on these two benchmarks is much wider: 85 vs 71 on SWE-bench between first and sixth place.
If someone quotes you a HumanEval score in 2026, ask them about SWE-bench instead.
Claude Opus 4.6 scores 79.3 on coding — now ahead of GPT-5.4 at 76.1. And Opus 4.6 is a non-reasoning model. It doesn't use chain-of-thought at inference time. GPT-5.4 does.
That means Opus is beating a reasoning model's coding output without the extra compute step. For latency-sensitive applications like autocomplete or interactive coding assistants, that matters. Reasoning models think before they respond, and that delay adds up when you're waiting for suggestions every few keystrokes.
If you need to self-host or fine-tune, Qwen3.5-122B-A10B currently leads the open-weight coding rows at 69.6, followed by Kimi K2.5 at 69.5, Qwen3.5-27B at 69.4, and GLM-4.7 at 67.6.
The gap between open-weight and the best proprietary coding rows is still real, but the picture is more nuanced than it was. The Qwen3.5 family now dominates the open-weight top spots, while GLM-5 at 66.6 is more balanced across coding, agentic, and math even if its pure coding row is lower.
Three standardized tests can't capture everything:
Multi-file refactors. No benchmark tests "rename this abstraction across 40 files and update all the tests."
Framework-specific knowledge. Does the model write idiomatic React, or code that works but looks like 2019?
Agent loop quality. How well a model performs when iterating — reading errors, retrying, editing files — doesn't show up in any single-pass benchmark.
Autocomplete and tab-completion: Use a non-reasoning model (Claude Opus 4.6, Grok 4.1). Faster responses, minimal quality difference at 1-5 line generation.
Debugging and bug fixes: GPT-5.4 Pro, Claude Opus 4.6, or Gemini 3.1 Pro depending on whether you want the strongest closed frontier row, the strongest non-reasoning row, or a strong balanced option.
Greenfield projects: Context window matters. Claude Opus 4.6, GPT-5.4 Pro, and Gemini 3.1 Pro all offer 1M tokens. GPT-5.3 Codex is limited to 400K.
Competitive programming: LiveCodeBench is still the benchmark. GPT-5.4 Pro leads at 86, GPT-5.3 Codex is at 85, GPT-5.4 is at 84, and GPT-5.2 is at 79.
→ See the full coding leaderboard · Full leaderboard
GPT-5.4 Pro currently wins the coding leaderboard, but Claude Opus 4.6 and Gemini 3.1 Pro are not far behind. Notably, Claude Opus 4.6 now leads GPT-5.4 (non-Pro) as a non-reasoning model. The more important takeaway is that coding rank is now much more sensitive to fresh repo-style engineering results than to saturated legacy benchmarks.
What is the best LLM for coding in 2026? GPT-5.4 Pro currently leads BenchLM's coding leaderboard at 88.3, followed by Claude Opus 4.6 at 79.3 and Gemini 3.1 Pro at 77.8.
Which coding benchmark should I use to compare LLMs? SWE-bench Verified and LiveCodeBench. HumanEval is saturated — six models score 91% and it no longer differentiates frontier models.
What is the best open source LLM for coding? Right now it is Qwen3.5-122B-A10B at 69.6 on BenchLM's coding score, followed by Kimi K2.5 at 69.5 and Qwen3.5-27B at 69.4 among open-weight rows.
Is Claude or GPT-5.4 better for coding? Claude Opus 4.6 now leads GPT-5.4 on BenchLM's current coding score, 79.3 to 76.1. The gap flipped from earlier snapshots, and Claude comes out on top as a non-reasoning model.
What do coding benchmarks miss? Multi-file refactors, framework-specific knowledge, agent loop quality, and IDE integration. SWE-bench and LiveCodeBench are the closest to real work, but no benchmark fully captures agentic coding in practice.
All benchmark data is from our coding leaderboard. Compare models on our comparison pages.
Coding benchmarks shift with every model release. We send one email a week with what moved and why.
Share This Report
Copy the link, post it, or save a PDF version.
On this page
Which models moved up, what’s new, and what it costs. One email a week, 3-min read.
Free. One email per week.
GPT-5.4 and Gemini 3.1 Pro are one point apart on BenchLM's leaderboard. We compare coding, math, reasoning, multimodal, agentic, pricing, and context window — with Gemini Deep Think as the reasoning wildcard.
State of LLM benchmarks in 2026: current BenchLM rankings, category leaders, benchmark trends, open vs closed performance, and what still matters after the latest scoring changes.
Which AI model is best for coding in 2026? We rank major LLMs by SWE-bench Pro and LiveCodeBench, with SWE-bench Verified shown as a historical baseline and React Native Evals tracked as a display benchmark for mobile app work.