Best Open Source LLMs (2026)

Last refreshed 2026-09-30. Next refresh: weekly.

The best open-weight LLMs in 2026, ranked by benchmark scores. Run locally, self-host, or deploy on your own infra — no API key required.

Release watch. Nemotron-Labs-Diffusion from NVIDIA, including 3B, 8B, and 14B diffusion language model variants.

Verdict

Use Kimi K3 for self-hosted open-weight use today.

Qwen3.8-Max is the runner-up, 0.9 points back on GPQA Diamond.

Researched 37d agoWhy this pickMethodology

How we rank

Open-weight boards emphasize GPQA Diamond (harder to game than broad MMLU), then MMLU, then recency.

  1. Eligibility — Models marked with supported open licenses/flags in seed data.
  2. Primary ranking — GPQA Diamond, then MMLU, then newer release.
  3. 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.
  4. Pricing — Hosted open-weight pricing still varies wildly — use the rate card column for apples-to-apples.
GPQA Diamond
#ModelInput $/1MOutput $/1M
1Kimi K3
ReasoningVisionTools

GPQA Diamond: 93.5%

$2.00$11.20
2MiniMax M3
ReasoningVisionTools

GPQA Diamond: 92.9%

$0.30$1.20
3Qwen3.8-Max
ReasoningVisionTools

GPQA Diamond: 92.6%

$2.00$6.00
4Qwen3.6-Max
Vision

GPQA Diamond: 91.8%

——
5Qwen3.8-Flash-Next
ReasoningVision

GPQA Diamond: 91.7%

$0.15$0.47
6GLM-5.2
ReasoningTools

GPQA Diamond: 91.2%

$1.40$4.40
7DeepSeek V4.1 Flash
ReasoningVisionTools

GPQA Diamond: 90.9%

$0.15$0.60
8Kimi K2.6
ReasoningVisionTools

GPQA Diamond: 90.5%

$0.73$3.40
9Qwen3.6-Plus
VisionTools

GPQA Diamond: 90.4%

$0.33$1.95
10DeepSeek V4 Pro
ReasoningTools

GPQA Diamond: 90.1%

$0.43$0.87
11Qwen3.5-397B-A17B
ReasoningVisionTools

GPQA Diamond: 89.3%

$0.39$2.34
12Qwen3.8-27B
ReasoningVisionTools

GPQA Diamond: 89.2%

$0.21$2.55
13Trinity-Large-Thinking
ReasoningTools

GPQA Diamond: 89.2%

$0.22$0.85
14Qwen3.5-Plus
Vision

GPQA Diamond: 88.4%

$0.30$1.80
15Ring-2.6-1T
ReasoningTools

GPQA Diamond: 88.27%

$0.07$0.63
16DeepSeek V4 Flash
ReasoningTools

GPQA Diamond: 88.1%

$0.06$0.12
17Qwen3.6-27B
ReasoningVisionTools

GPQA Diamond: 87.8%

$0.32$2.70
18DeepSeek V3 0324

GPQA Diamond: 87.6%

$0.27$1.12
19MiniMax M2.7
ReasoningTools

GPQA Diamond: 87.4%Tied within margin

$0.28$1.20
20Hunyuan Hy3 Preview
PreviewReasoningTools

GPQA Diamond: 87.2%

$0.07$0.26

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.
  • #4GLM-5.2

    GLM-5.2 is Z.ai's coding-first successor to GLM-5.1 in the GLM-5 family, released June 13 2026. 753B parameters (40B active) in IndexShare MoE architecture; the IndexShare innovation reuses the same attention indexer across every four sparse layers, cutting per-token FLOPs by 2.9x at 1M context length. Trained on 28.5T tokens. Supports a 1M-token context window via the glm-5.2[1m] model ID, with 131,072-token maximum output and High/Max thinking-effort levels designed for extended agentic coding sessions. MIT license; open weights available on Hugging Face (zai-org/GLM-5.2 and zai-org/GLM-5.2-FP8). Self-reported HF card benchmarks: SWE-bench Pro 62.1, Terminal-Bench 2.1 82.7, MCP-Atlas 76.8, Tool-Decathlon 48.2, GPQA Diamond 91.2, AIME 2026 99.2, HLE 40.5. Available to GLM Coding Plan subscribers (Lite/Pro/Max/Team) directly, and via OpenRouter token API ($1.40/$4.40 per 1M tokens).

    91.2%

    GPQA Diamond

  • DeepSeek V4.1 Flash is a 552B-parameter (8B prefill / 16B decode activated) mixture-of-experts model with Compressed Expert Dispatch (CED), released September 10, 2026 under the MIT license. It supports 1M-token context, up to 384K output tokens, multimodal vision input, reasoning, function calling, tool use, structured outputs, and prompt caching. Weights are available on Hugging Face. The DeepSeek API serves it as deepseek-flash; legacy API aliases deepseek-v4-flash and deepseek-v4-flash-vision-exp route to this model. Off-peak API pricing: $0.15/1M input, $0.60/1M output (cache read: $0.003/1M); peak hours are 2×.

    90.9%

    GPQA Diamond

  • #6Kimi K2.6

    Kimi K2.6 is Moonshot AI's multimodal agentic coding model, released April 20 2026 under a Modified MIT license. Built on a 1-trillion-parameter MoE architecture (32B active, 384 experts with 8 selected per token plus 1 shared expert, 61 layers), it features a 262K context window and up to 65,536 output tokens. Supports native image and video inputs (screenshots, PDFs, spreadsheets). Designed for long-horizon coding with agent swarms of up to 300 sub-agents and 4,000 coordinated steps; Moonshot AI cites 200–300 sequential tool calls without task drift. Key benchmarks: SWE-bench Verified 80.2%, SWE-bench Pro 58.6%, LiveCodeBench v6 89.6%, GPQA Diamond 90.5%, Terminal-Bench 2.0 66.7%. Chatbot Arena Elo 1454 (2026-04-28 snapshot).

    90.5%

    GPQA Diamond

Compare Top Picks

Side-by-side comparison of the top picks by price, benchmark, and API access.
Kimi K3 vs MiniMax M3Kimi K3 vs Qwen3.8-MaxKimi K3 vs Qwen3.6-MaxKimi K3 vs Qwen3.8-Flash-NextMiniMax M3 vs Qwen3.8-MaxMiniMax M3 vs Qwen3.6-Max

Browse Other Categories

Best LLMs for Code GenerationBest LLMs for RAGBest LLMs for Agents vs Agent Products (2026)Best LLMs for ClassificationBest 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