1stTop pick
Researched 8d agoGemini 3 Pro
- SWE-bench Verified
- 76.2%
- Output (from)
- $5.00 / 1M
Best LLMs for Code Generation (2026)
Last refreshed 2026-09-29. Next refresh: weekly.
The top coding LLMs in 2026, ranked by SWE-bench and HumanEval. Includes API pricing and context window for each pick — updated daily.
Verdict
Use Gemini 3 Pro for code generation today.
Phi-3 Mini 128K is the runner-up: 76.2% vs — on SWE-bench Verified.
2ndShortlist
Researched 8d agoPhi-3 Mini 128K
- SWE-bench Verified
- —
- Output (from)
- $0.100 / 1M
3rdShortlist
Researched 8d agoQwen2.5-14B
- SWE-bench Verified
- —
- Output (from)
- $0.200 / 1M
How we rank
Coding leaders are ordered on shipped coding-agent evidence first, then classic code generation scores, with recency as the last tie-break.
- Eligibility — Chat models tied to code work with current public/self-serve availability: code specialization, tracked code-execution flags, scores on SWE-bench / HumanEval / LiveCodeBench / Aider / BigCodeBench, or known code-family slugs.
- Primary ranking — SWE-bench Verified (higher is better), then HumanEval, then SWE-bench Pro.
- Benchmark variants — Independent standardized rows such as Vals.ai, Scale AI SWE-bench Pro, and Nebius/OpenHands SWE-bench Verified stay separate from vendor-reported variants. Vendor-only orchestration claims do not promote a model on this page until a comparable public harness publishes the score.
- Tie-breaks — Newer `release` date when benchmark scores match.
- Variant collapse — We keep one row per model family (`familySlug` + parameter tier). When headline scores tie within ±0.5 pt (±10 Elo on Chatbot Arena), we pick the canonical SKU by lowest tracked input price, then GA over preview or limited access, then newest `release`. A folded sibling within the benchmark noise band can show a "Tied within margin" chip on that score cell.
- Preview fallback — The podium prefers GA models. Preview or invite-only candidates can fill this page only when fewer than three GA coding primaries remain after the gate.
- Pricing column — Input/output prices are the lowest tracked public commercial rate cards in seed data; partner-only pricing is kept out of the default ranking.
New models awaiting benchmark coverage
These source-backed rows qualify for this task page, but they are not scored leaderboard picks until the category benchmark data exists.
| Model | Why it is listed | Status | Tracked price |
|---|---|---|---|
| GPT-6.1 Sol ToolsCode execution | GPT-6.1 Sol is a newly researched coding-capable model; keep it on the watchlist until category scores land. | Benchmark pending No tracked SWE-bench Verified, HumanEval, or SWE-bench Pro score yet. | In $2.00 / Out $10.00 |
| GPT-6 Luna ToolsCode execution | GPT-6 Luna is a newly researched coding-capable model; keep it on the watchlist until category scores land. | Benchmark pending No tracked SWE-bench Verified, HumanEval, or SWE-bench Pro score yet. | In $0.10 / Out $0.50 |
| GPT-6 Sol ToolsCode execution | GPT-6 Sol is a newly researched coding-capable model; keep it on the watchlist until category scores land. | Benchmark pending No tracked SWE-bench Verified, HumanEval, or SWE-bench Pro score yet. | In $2.00 / Out $10.00 |
| Grok 4.7 ToolsCode execution | Grok 4.7 is a newly researched coding-capable model; keep it on the watchlist until category scores land. | Benchmark pending No tracked SWE-bench Verified, HumanEval, or SWE-bench Pro score yet. | In $1.60 / Out $4.80 |
| # | Model | Input $/1M | Output $/1M | |
|---|---|---|---|---|
| 1 | Claude Fable 5 ReasoningVisionTools SWE-bench Verified: 96% | $10.00 | $50.00 | |
| 2 | Claude Opus 5 ReasoningVisionTools SWE-bench Verified: 96% | $5.00 | $25.00 | |
| 3 | Claude Opus 4.8 ReasoningVisionTools SWE-bench Verified: 88.6% | $5.00 | $25.00 | |
| 4 | Claude Opus 4.7 ReasoningVisionTools SWE-bench Verified: 87.6% | $5.00 | $25.00 | |
| 5 | Claude Sonnet 5 ReasoningVisionTools SWE-bench Verified: 85.2% | $2.00 | $10.00 | |
| 6 | GPT-5.3-Codex ReasoningVisionTools SWE-bench Verified: 85% | $1.75 | $14.00 | |
| 7 | GPT-5.5 ReasoningVisionTools SWE-bench Verified: 82.6% | $5.00 | $30.00 | |
| 8 | GPT-5.5 Pro ReasoningVisionTools SWE-bench Verified: 82.6% | $30.00 | $180.00 | |
| 9 | Claude Opus 4.5 ReasoningVisionTools SWE-bench Verified: 80.9% | $5.00 | $25.00 | |
| 10 | Claude Opus 4.6 ReasoningVisionTools SWE-bench Verified: 80.8% | $5.00 | $25.00 | |
| 11 | Gemini 3.1 Pro Preview PreviewVisionTools SWE-bench Verified: 80.6% | $2.00 | $12.00 | |
| 12 | DeepSeek V4 Pro ReasoningTools SWE-bench Verified: 80.6% | $0.43 | $0.87 | |
| 13 | MiniMax M3 ReasoningVisionTools SWE-bench Verified: 80.5% | $0.30 | $1.20 | |
| 14 | Qwen3.7-Max ReasoningTools SWE-bench Verified: 80.4% | $1.25 | $3.75 | |
| 15 | Kimi K2.6 ReasoningVisionTools SWE-bench Verified: 80.2% | $0.73 | $3.40 | |
| 16 | MiniMax M2.5 Highspeed ReasoningTools SWE-bench Verified: 80.2% | $0.60 | $2.40 | |
| 17 | Claude Sonnet 4.6 ReasoningVisionTools SWE-bench Verified: 79.6% | $3.00 | $15.00 | |
| 18 | DeepSeek V4 Flash ReasoningTools SWE-bench Verified: 79% | $0.06 | $0.12 | |
| 19 | Xiaomi MiMo-V2.5-Pro Tools SWE-bench Verified: 78.9% | $0.43 | $0.87 | |
| 20 | Qwen3.6 Max Preview PreviewReasoningVisionTools SWE-bench Verified: 78.8% | $1.04 | $6.24 |
Honorable mentions
Next seats in this ranking. Lines below are from each model's stored description in LLMReference seed data—spot-check the model page before relying on a capability claim.
Aya-23-8B is a multilingual large language model developed by Cohere For AI, featuring 8 billion parameters. As an instruction-fine-tuned model, it is adept at following instructions and is optimized for text generation and understanding. The model employs a decoder-only Transformer architecture, utilizing enhancements like parallel attention and feed-forward layers for efficiency. It supports 23 languages, including Arabic, Chinese, English, and French, and is proficient in tasks such as machine translation, chatbot interactions, and text summarization. Despite its capabilities, its performance might vary across languages, particularly those with less linguistic resources, and it has a context length limit of 8192 tokens. Training involved diverse data sources like human annotations and synthetic datasets to bolster its multilingual proficiency.
—
SWE-bench Verified
Instruction-tuned 7B variant combining strong reasoning with real-time inference on single GPUs, ideal for developer tools and vision applications.
—
SWE-bench Verified
The Llama 3 8B Instruct model, released on April 18, 2024, is Meta's latest instruction-following language model with 8 billion parameters. It utilizes an auto-regressive transformer architecture with Grouped-Query Attention for improved scalability. Trained on over 15 trillion tokens and fine-tuned with 10 million human-annotated examples, it excels in dialogue and conversational tasks. The model outperforms its predecessors on industry benchmarks, scoring 68.4 on MMLU (5-shot). Designed for commercial and research applications, it prioritizes safety and helpfulness, making it suitable for chatbots, virtual assistants, and other interactive AI applications. For more details, visit the Hugging Face page [1].
—
SWE-bench Verified
Compare Top Picks
Side-by-side comparison of the top picks by price, benchmark, and API access.
Browse Other Categories
Best LLMs for RAGBest LLMs for Agents vs Agent Products (2026)Best LLMs for ClassificationBest Open Source LLMsBest Multimodal / Vision LLMsBest LLM for Translation in 2026Best AI Image Models in 2026Best AI Video Models in 2026Best LLMs for Reasoning & MathBest Small Language Models (SLMs)Best LLMs for JSON / Tool useCheapest LLM APIs You Can Call Right NowBest Long Context LLMsBest Mainstream LLM APIs, RankedBest LLMs for EnterpriseBest Free LLMs You Can Use Right NowBest LLMs for WritingBest LLMs for MarketingBest LLMs for Customer Support