Skip to content
OpenKey

Best LLMs for coding

Models with head-to-head Design Arena results on code tasks. 94 of 338 catalog models match as of July 2026.

94
Models
$0.031
Input from /1M
10M
Max context
Jun 16, 2026
Newest release

94 models in this catalog are viable for coding work, priced anywhere from $0 (three free entries, including Nemotron 3 Ultra free) up to $15/$75 per M tokens for Claude Opus 4.1. The newest addition is GLM 5.2 (June 16, 2026), running a 1,048,576-token context window at $0.93/$3.00 per M and ranked #1 on design_arena's 3D benchmark with an Elo of 1376. Claude Opus 4.8 leads androidnative generation (rank 1, Elo 1343), and DeepSeek V4 Pro delivers 3D rank 7 (Elo 1327) at $0.435/$0.87 — proof that frontier scores don't require frontier prices. If you're picking a model for a real coding workload, the spread between cheapest paid and most expensive paid option is roughly 500x on input price alone, so the tradeoffs below matter more than any single leaderboard slot.

Top picks

GLM 5.2

Newest model in the set (2026-06-16), 1,048,576-token context, and rank 1 on design_arena's 3D benchmark (Elo 1376) at $0.93 input / $3.00 output per M tokens — on OpenKey that's $0.958/$3.09 after the 3% fee. Good default if you want current-generation quality without paying Opus-tier prices.

DeepSeek V4 Pro

$0.435 input / $0.87 output per M ($0.448/$0.896 on OpenKey) with a 3D rank of 7 (Elo 1327) — within 50 Elo points of the category leader for a fraction of the cost. Text-only, 1,048,576-token context, released 2026-04-24.

Claude Opus 4.8

Rank 1 on androidnative generation (Elo 1343) with full image+file input support. At $5/$25 per M ($5.15/$25.75 on OpenKey) it's not cheap, but it's the top pick when the coding task involves multimodal context and mobile-native code output specifically.

All 94 models

Full catalog →
  • GLM 5.2ctx 1.0M · inout
  • Kimi K2.7 Codectx 262K · inout
  • Claude Fable 5ctx 1M · inout
  • Nex-N2-Proctx 262K · inout
  • Nemotron 3 Ultractx 1M · inout
  • Nemotron 3 Ultra (free)Freectx 1M · inFreeoutFree
  • MiniMax M3ctx 1.0M · inout
  • Step 3.7 Flashctx 256K · inout
  • Claude Opus 4.8ctx 1M · inout
  • Qwen3.7 Maxctx 1M · inout
  • Gemini 3.5 Flashctx 1.0M · inout
  • Grok 4.3ctx 1M · inout
  • DeepSeek V4 Flashctx 1.0M · inout
  • DeepSeek V4 Proctx 1.0M · inout
  • GPT-5.5ctx 1.1M · inout
  • MiMo-V2.5ctx 1.0M · inout
  • MiMo-V2.5-Proctx 1.0M · inout
  • Kimi K2.6ctx 262K · inout
  • Claude Opus 4.7ctx 1M · inout
  • GLM 5.1ctx 203K · inout
  • Qwen3.6 Plusctx 1M · inout
  • Trinity Large Thinkingctx 262K · inout
  • GLM 5V Turboctx 203K · inout
  • MiniMax M2.7ctx 205K · inout
  • GLM 5 Turboctx 262K · inout
  • GPT-5.4ctx 1.1M · inout
  • Mercury 2ctx 128K · inout
  • Gemini 3.1 Flash Lite Previewctx 1.0M · inout
  • GPT-5.3-Codexctx 400K · inout
  • Gemini 3.1 Pro Previewctx 1.0M · inout
  • Claude Sonnet 4.6ctx 1M · inout
  • Qwen3.5 397B A17Bctx 256K · inout
  • Qwen3.5 Plus 2026-02-15ctx 1M · inout
  • MiniMax M2.5ctx 205K · inout
  • GLM 5ctx 203K · inout
  • Claude Opus 4.6ctx 1M · inout
  • Kimi K2.5ctx 262K · inout
  • GLM 4.7 Flashctx 203K · inout
  • MiniMax M2.1ctx 205K · inout
  • GLM 4.7ctx 203K · inout
  • Gemini 3 Flash Previewctx 1.0M · inout
  • GPT-5.2ctx 400K · inout
  • Ministral 3 14B 2512ctx 262K · inout
  • Ministral 3 3B 2512ctx 131K · inout
  • Ministral 3 8B 2512ctx 262K · inout
  • DeepSeek V3.2ctx 131K · inout
  • Mistral Large 3 2512ctx 262K · inout
  • Claude Opus 4.5ctx 200K · inout
  • GPT-5.1ctx 400K · inout
  • GPT-5.1-Codexctx 400K · inout
  • GPT-5.1-Codex-Minictx 400K · inout
  • MiniMax M2ctx 205K · inout
  • Claude Haiku 4.5ctx 200K · inout
  • GLM 4.6ctx 203K · inout
  • Claude Sonnet 4.5ctx 1M · inout
  • DeepSeek V3.2 Expctx 164K · inout
  • Gemini 2.5 Flash Lite Preview 09-2025ctx 1.0M · inout
  • Qwen3 Maxctx 262K · inout
  • DeepSeek V3.1 Terminusctx 164K · inout
  • Kimi K2 0905ctx 262K · inout
  • DeepSeek V3.1ctx 164K · inout
  • Mistral Medium 3.1ctx 131K · inout
  • GPT-5ctx 400K · inout
  • GPT-5 Minictx 400K · inout
  • GPT-5 Nanoctx 400K · inout
  • Claude Opus 4.1ctx 200K · inout
  • gpt-oss-120bctx 131K · inout
  • gpt-oss-120b (free)Freectx 131K · inFreeoutFree
  • Codestral 2508ctx 256K · inout
  • Qwen3 235B A22B Thinking 2507ctx 262K · inout
  • GLM 4.5ctx 131K · inout
  • GLM 4.5 Airctx 131K · inout
  • Qwen3 Coder 480B A35Bctx 1.0M · inout
  • Qwen3 Coder 480B A35B (free)Freectx 1.0M · inFreeoutFree
  • Qwen3 235B A22B Instruct 2507ctx 262K · inout
  • Kimi K2 0711ctx 131K · inout
  • Mistral Small 3.2 24Bctx 128K · inout
  • Gemini 2.5 Flashctx 1.0M · inout
  • Gemini 2.5 Proctx 1.0M · inout
  • R1 0528ctx 164K · inout
  • Claude Opus 4ctx 200K · inout
  • Claude Sonnet 4ctx 1M · inout
  • Mistral Medium 3ctx 131K · inout
  • Qwen3 235B A22Bctx 131K · inout
  • Qwen3 30B A3Bctx 131K · inout
  • o3ctx 200K · inout
  • o4 Minictx 200K · inout
  • GPT-4.1ctx 1.0M · inout
  • GPT-4.1 Minictx 1.0M · inout
  • GPT-4.1 Nanoctx 1.0M · inout
  • Llama 4 Maverickctx 1.0M · inout
  • Llama 4 Scoutctx 10M · inout
  • DeepSeek V3ctx 131K · inout
  • GPT-4octx 128K · inout

Prices per 1M tokens, flat 3% fee included. Hover or tap a price for the math.

How to choose

First tradeoff: benchmark specialization vs. general coding use. This catalog's benchmarks are design_arena categories (3D, SVG, webapps, gamedev, androidnative, etc.) — a model ranked 1 on androidnative (Claude Opus 4.8) isn't necessarily your best pick for SVG generation, where Gemini 3.1 Pro Preview sits at rank 2 (Elo 1347) for $2/$12 per M. Match the benchmark to your actual task. Second: price-to-context tradeoff. DeepSeek V4 Flash is $0.09/$0.18 per M with a 1,048,576-token window, cheapest by far among frontier-context models, but its best benchmark rank (svg, 23rd, Elo 1216) trails the leaders. Third: modality. If your coding workflow needs image or file input (screenshots, diagrams, PDFs), you're limited to text+image or text+image+file models like Claude Opus 4.8 or GPT-5.5 — pure text->text models like GLM 5.2 or DeepSeek V4 Pro won't take that input regardless of benchmark score.

Questions

What's the cheapest model in this coding list?
Among paid models, gpt-oss-120b is cheapest at $0.03 input / $0.15 output per M tokens ($0.0309/$0.1545 on OpenKey after the 3% fee). There are also 3 free models in the set, including Nemotron 3 Ultra (free), which carries the same 1,000,000-token context and benchmark rank as its paid version but at $0 cost.
Is there a free option that's actually competitive?
Nemotron 3 Ultra (free) mirrors the paid Nemotron 3 Ultra exactly — same 1,000,000-token context, same design_arena 3D rank of 40 (Elo 1213). It won't top any leaderboard, but for zero cost it's a reasonable starting point before you commit to a paid model.
How much more does the top-ranked model cost versus a mid-tier one?
GLM 5.2 (3D rank 1, Elo 1376) runs $0.93/$3.00 per M. DeepSeek V4 Pro (3D rank 7, Elo 1327) runs $0.435/$0.87 per M — less than half the price for a 49-point Elo gap. Whether that gap matters depends on how sensitive your use case is to the top few ranks versus the top 10.