Best Open Source LLMs (2026)

As of 2026-11-14, LLM Reference's top pick for self-hosted open-weight use is Kimi K3.1 Pro by Moonshot AI (GPQA Diamond 94.2%, $10.00/1M output tokens), chosen from models that meet the podium eligibility rules. Second pick: DeepSeek V4.1 Flash 1017 by DeepSeek.

Last refreshed 2026-11-14. 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 K3.1 Pro for self-hosted open-weight use today.

DeepSeek V4.1 Flash 1017 is the runner-up, 1 point back on GPQA Diamond.

Researched 11d 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. Podium eligibility — Preview, restricted, retired and retiring models never fill a podium slot. Preview means preview or invite-only access, or a model that is announced or pre-release. Restricted means restricted, partner, private or otherwise unavailable access. Retired means the model is past its retirement date or every provider route is retired. Retiring means the model, or any one of its provider routes, has a retirement date within the next 60 days. These models can still appear in the table below the podium.
  5. Pricing — Hosted open-weight pricing still varies wildly — use the rate card column for apples-to-apples.
#ModelInput $/1MOutput $/1M
1Kimi K3.1 Pro
ReasoningVisionTools

GPQA Diamond: 94.2%

$2.00$10.00
2DeepSeek V4.1 Flash 1017
ReasoningVisionTools

GPQA Diamond: 93.18%

$0.20$0.60
3Kimi K3
ReasoningVisionTools

GPQA Diamond: 93.12%

$3.00$15.00
4Qwen3.8-Max
ReasoningVisionTools

GPQA Diamond: 92.6%

$2.00$6.00
5GLM-5.2
ReasoningTools

GPQA Diamond: 91.86%

$1.40$4.40
6Qwen3.6-Max
Vision

GPQA Diamond: 91.8%

——
7Qwen3.8-Flash-Next
ReasoningVisionVerify pricing

GPQA Diamond: 91.7%

——
8MiniMax M3
ReasoningVisionToolsVerify pricing

GPQA Diamond: 90.91%

$0.30$1.20
9Beam
ReasoningTools

GPQA Diamond: 90.6%

$1.00$3.00
10Qwen3.6-Plus
VisionTools

GPQA Diamond: 90.4%

$0.33$1.95
11GLM-5.3-Flash
ReasoningVisionTools

GPQA Diamond: 90.15%

$0.15$0.50
12DeepSeek V4 Pro
ReasoningToolsVerify pricing

GPQA Diamond: 90.1%

$0.36$0.73
13Nanbeige4.3-33B
ReasoningTools

GPQA Diamond: 89.5%

——
14Qwen3.8-27B
ReasoningVisionToolsVerify pricing

GPQA Diamond: 89.2%

$0.42$2.55
15Trinity-Large-Thinking
ReasoningToolsVerify pricing

GPQA Diamond: 89.2%

$0.25$0.80
16Muse Spark Open
ReasoningVisionTools

GPQA Diamond: 89%

$1.25$4.25
17Nemotron 3.5 Bolt 74B-A9B
ReasoningVisionTools

GPQA Diamond: 88.83%

——
18Inkling-Small
ReasoningVision

GPQA Diamond: 88.5%

$0.45$1.20
19Qwen3.5-Plus
Vision

GPQA Diamond: 88.4%

$0.30$1.80
20Inkling
ReasoningVisionVerify pricing

GPQA Diamond: 88.3%

$1.00$4.05

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.
  • Qwen3.8-Max is a 2.4-trillion-parameter MoE flagship delivering a comprehensive leap in coding and professional work. It autonomously codes and delivers complete projects spanning 10+ days, handles hundreds of specialized tasks across legal, financial, design, and other professional domains, and produces production-grade results end-to-end in a single conversation. Native visual understanding runs through the full cycle of planning, execution, and verification, enabling deep semantic analysis of ultra-long documents and extended video content. The public Hugging Face checkpoint (Qwen3.8-2.4T-A95B) is text-only under the Qwen3.8-Max License; the hosted qwen3.8-max API adds vision, optional non-thinking mode, 1M context, tools, structured outputs, and prompt caching on the same model ID.

    92.6%

    GPQA Diamond

  • 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).

    91.86%

    GPQA Diamond

  • Qwen3.6-Max builds upon Qwen3-Max and Qwen3.6-Plus to deliver enhanced vibe coding capabilities, more efficient coding agent execution, and significantly improved front-end development skills. Its long-tail knowledge retention has been further upgraded, making it Alibaba's most capable multimodal flagship as of April 2026.

    91.8%

    GPQA Diamond

Compare Top Picks

Side-by-side comparison of the top picks by price, benchmark, and API access.

Frequently asked questions

What is the best LLM for self-hosted open-weight use?

LLM Reference's top pick for self-hosted open-weight use as of 2026-11-14 is Kimi K3.1 Pro by Moonshot AI, with GPQA Diamond 94.2%. Picks come from models that meet the podium eligibility rules, so the top pick may not lead the full table. Second pick: DeepSeek V4.1 Flash 1017 by DeepSeek (GPQA Diamond 93.18%).

Which ranked LLM for self-hosted open-weight use has the lowest output price?

Among the ranked models shown on this page, GLM-5.3-Flash by Zhipu AI has the lowest listed output price: $0.500/1M output tokens, at rank 11.

What is the best open-weight LLM for self-hosted open-weight use?

Kimi K3.1 Pro by Moonshot AI is the highest-ranked open-weight model shown on this page (rank 1, GPQA Diamond 94.2%). License: Kimi K3.1 Pro License.

How is the self-hosted open-weight use ranking calculated?

Open-weight boards emphasize GPQA Diamond (harder to game than broad MMLU), then MMLU, then recency. Preview, restricted, retired and retiring models never fill a podium slot. Preview means preview or invite-only access, or a model that is announced or pre-release. Restricted means restricted, partner, private or otherwise unavailable access. Retired means the model is past its retirement date or every provider route is retired. Retiring means the model, or any one of its provider routes, has a retirement date within the next 60 days. These models can still appear in the table below the podium.

How often is this page updated?

Weekly. Last refreshed 2026-11-14.