# Factuality and anti-hallucination leaderboard

As of November 21, 2026, Gemini 4 Carbon leads our factuality and anti-hallucination leaderboard at 78.8 (Supported, $2/$10 per 1M tokens), ahead of GPT-6 Astra at 77.9. Gemini 3.8 Flash (#6, 72.6) is the lowest-cost model near the top at $0.75/$3.75 per 1M tokens; DeepSeek V4.1 Flash 1017 leads open-weight models at #18 (61.9). We rank 65 models (64 Supported, 1 Estimated) across 10 domain protocols, and this vertical carries zero weight in the overall score.

## Shortlist by workload and budget

- **Highest point score:** [Gemini 4 Carbon](/models/gemini-4-carbon) (#1, 78.8, Supported, $2/$10 per 1M tokens) — Costs $2/$10 per 1M tokens; Gemini 3.8 Flash (#6, 72.6) cuts token spend to $0.75/$3.75 per 1M tokens.
- **Closest contender:** [GPT-6 Astra](/models/gpt-6-astra) (#2, 77.9, Supported, $10/$50 per 1M tokens) — Sits 0.9 points behind Gemini 4 Carbon; test both on your own prompts when provider lock-in or latency matters.
- **Best API value:** [Gemini 3.8 Flash](/models/gemini-3-8-flash) (#6, 72.6, Supported, $0.75/$3.75 per 1M tokens) — Keeps 92% of the top score (72.6 vs 78.8), but gives up 6.2 points on the hardest tail cases.
- **Top open-weight model:** [DeepSeek V4.1 Flash 1017](/models/deepseek-v4-1-flash) (#18, 61.9, Supported, Open weights · $0.3/$1.2 per 1M tokens) — Trails Gemini 4 Carbon by 16.9 points (#18 overall here), so pick it when weight control or on-prem residency outweighs peak score.

## Ranked models

| Rank | Model | Creator | Score | Conditional range | Evidence | API ($/1M in/out) | Context | Source families |
|---:|---|---|---:|---:|---|---:|---:|---:|
| #1 | [Gemini 4 Carbon](/models/gemini-4-carbon) | Google | 78.8 | 68.5-89.0 | Supported | $2.00 / $10.00 | 1M | 4 |
| #2 | [GPT-6 Astra](/models/gpt-6-astra) | OpenAI | 77.9 | 68.0-87.7 | Supported | $10.00 / $50.00 | 1.05M | 6 |
| #3 | [Gemini 4 Argon](/models/gemini-4-argon) | Google | 77.1 | 67.5-86.8 | Supported | $2.00 / $10.00 | 1M | 6 |
| #4 | [Gemini 4 Hydrogen](/models/gemini-4-hydrogen) | Google | 76.9 | 66.0-87.8 | Supported | $1.50 / $7.50 | 1M | 2 |
| #5 | [GPT-6.1 Sol](/models/gpt-6-1-sol) | OpenAI | 76.4 | 66.6-86.1 | Supported | $2.00 / $10.00 | 1.05M | 6 |
| #6 | [Gemini 3.8 Flash](/models/gemini-3-8-flash) | Google | 72.6 | 62.8-82.4 | Supported | $0.75 / $3.75 | 1M | 6 |
| #7 | [GPT-6.1 Terra](/models/gpt-6-1-terra) | OpenAI | 72.3 | 62.1-82.6 | Supported | $1.00 / $5.00 | 1.05M | 4 |
| #8 | [Gemini 3.1 Pro](/models/gemini-3-1-pro) | Google | 72.0 | 62.2-81.9 | Supported | $2.00 / $12.00 | 1M | 6 |
| #9 | [GPT-6 Sol](/models/gpt-6-sol) | OpenAI | 71.7 | 61.9-81.5 | Supported | $2.00 / $10.00 | 1.05M | 6 |
| #10 | [GPT-5.6 Sol](/models/gpt-5-6-sol) | OpenAI | 71.2 | 61.5-81.0 | Supported | $4.00 / $20.00 | 1.05M | 7 |
| #11 | [Gemini 3.7 Flash](/models/gemini-3-7-flash) | Google | 70.9 | 60.9-80.9 | Supported | $0.75 / $3.75 | 1M | 6 |
| #12 | [Gemini 3 Pro](/models/gemini-3-pro) | Google | 69.9 | 58.5-81.3 | Supported | $2.00 / $12.00 | 2M | 3 |
| #13 | [Gemini 3.6 Flash](/models/gemini-3-6-flash) | Google | 68.0 | 58.1-77.9 | Supported | $0.75 / $3.75 | 1M | 5 |
| #14 | [Claude Opus 5.5](/models/claude-opus-5-5) | Anthropic | 66.3 | 56.3-76.3 | Supported | $4.00 / $20.00 | 1M | 4 |
| #15 | [Claude Opus 5](/models/claude-opus-5) | Anthropic | 65.9 | 55.9-76.0 | Supported | $5.00 / $25.00 | - | 5 |
| #16 | [Gemini 3.5 Flash](/models/gemini-3-5-flash) | Google | 64.7 | 54.5-74.8 | Supported | $1.50 / $9.00 | 1M | 5 |
| #17 | [GPT-6 Luna](/models/gpt-6-luna) | OpenAI | 64.6 | 54.5-74.7 | Supported | $0.10 / $0.50 | 1.05M | 4 |
| #18 | [DeepSeek V4.1 Flash 1017](/models/deepseek-v4-1-flash) | DeepSeek | 61.9 | 51.5-72.3 | Supported | $0.30 / $1.20 | 1M | 3 |
| #19 | [GPT-5.5](/models/gpt-5-5) | OpenAI | 61.3 | 45.7-76.9 | Supported | $5.00 / $30.00 | 1M | 2 |
| #20 | [Claude Opus 4.8](/models/claude-opus-4-8) | Anthropic | 61.3 | 51.3-71.3 | Supported | $5.00 / $25.00 | 1M | 5 |
| #21 | [GPT-5.6 Terra](/models/gpt-5-6-terra) | OpenAI | 61.1 | 51.3-70.9 | Supported | $2.00 / $12.00 | 1.05M | 7 |
| #22 | [Claude Opus 4.7](/models/claude-opus-4-7) | Anthropic | 61.0 | 45.2-76.9 | Supported | $5.00 / $25.00 | 1M | 2 |
| #23 | [Grok 4.6](/models/grok-4-6) | xAI | 60.4 | 50.7-70.1 | Supported | $2.00 / $6.00 | 500K | 6 |
| #24 | [Grok 4.5](/models/grok-4-5) | xAI | 59.3 | 49.4-69.2 | Supported | $2.00 / $6.00 | 500K | 6 |
| #25 | [GLM-5.3](/models/glm-5-3) | Z.AI | 59.0 | 48.3-69.8 | Supported | Free | 1M | 3 |
| #26 | [Claude Opus 4.6](/models/claude-opus-4-6) | Anthropic | 58.9 | 48.3-69.5 | Supported | $5.00 / $25.00 | 1M | 4 |
| #27 | [Claude Opus 4.6 (Adaptive)](/models/claude-opus-4-6-thinking) | Anthropic | 58.9 | 48.3-69.5 | Supported | - | 1M | 4 |
| #28 | [Grok 4](/models/grok-4) | xAI | 58.8 | 46.6-71.1 | Supported | - | 128K | 3 |
| #29 | [Claude Sonnet 5.5](/models/claude-sonnet-5-5) | Anthropic | 58.5 | 48.7-68.3 | Supported | $2.00 / $10.00 | 1M | 6 |
| #30 | [o3](/models/o3) | OpenAI | 58.1 | 46.2-69.9 | Supported | $2.00 / $8.00 | 200K | 4 |
| #31 | [GPT-5.6 Luna](/models/gpt-5-6-luna) | OpenAI | 57.2 | 47.4-67.0 | Supported | $0.20 / $1.20 | 1.05M | 6 |
| #32 | [Qwen3.8 Max](/models/qwen3-8-max) | Alibaba | 57.2 | 47.0-67.3 | Supported | - | 1M | 4 |
| #33 | [Claude Opus 4.5](/models/claude-opus-4-5) | Anthropic | 55.6 | 44.0-67.2 | Supported | $5.00 / $25.00 | 200K | 3 |
| #34 | [Claude Opus 4.5 Thinking](/models/claude-opus-4-5-thinking) | Anthropic | 55.6 | 44.0-67.2 | Supported | - | 200K | 3 |
| #35 | [GPT-5.2](/models/gpt-5-2) | OpenAI | 54.9 | 39.4-70.5 | Supported | $1.75 / $14.00 | 400K | 2 |
| #36 | [Gemini 3.5 Flash-Lite](/models/gemini-3-5-flash-lite) | Google | 54.6 | 44.6-64.6 | Supported | $0.30 / $2.50 | 1M | 6 |
| #37 | [GPT-5.4](/models/gpt-5-4) | OpenAI | 53.1 | 42.4-63.8 | Supported | $2.50 / $15.00 | 1.05M | 4 |
| #38 | [GPT-5.1](/models/gpt-5-1) | OpenAI | 52.8 | 40.4-65.3 | Supported | $1.25 / $10.00 | 200K | 3 |
| #39 | [Claude Sonnet 5](/models/claude-sonnet-5) | Anthropic | 51.8 | 41.6-62.0 | Supported | $2.00 / $10.00 | 1M | 4 |
| #40 | [Claude Sonnet 4.6](/models/claude-sonnet-4-6) | Anthropic | 51.6 | 40.7-62.4 | Supported | $3.00 / $15.00 | 200K | 4 |
| #41 | [GPT-5](/models/gpt-5) | OpenAI | 49.7 | 34.6-64.8 | Supported | $1.25 / $10.00 | 128K | 3 |
| #42 | [Gemini 2.5 Flash](/models/gemini-2-5-flash) | Google | 48.9 | 35.6-62.1 | Supported | $0.30 / $2.50 | 1M | 4 |
| #43 | [DeepSeek-R1](/models/deepseek-r1) | DeepSeek | 47.5 | 32.8-62.1 | Supported | $0.55 / $2.19 | 128K | 4 |
| #44 | [DeepSeek V3](/models/deepseek-v3) | DeepSeek | 47.2 | 32.4-61.9 | Supported | $0.27 / $1.10 | 128K | 3 |
| #45 | [Claude Haiku 5.5](/models/claude-haiku-5-5) | Anthropic | 46.7 | 36.9-56.5 | Supported | $0.10 / $0.50 | 1M | 6 |
| #46 | [Claude Sonnet 4.5](/models/claude-sonnet-4-5) | Anthropic | 46.4 | 34.4-58.5 | Supported | $3.00 / $15.00 | 200K | 3 |
| #47 | [Claude Sonnet 4.5 Thinking](/models/claude-sonnet-4-5-thinking) | Anthropic | 46.4 | 34.4-58.5 | Supported | - | 200K | 3 |
| #48 | [DeepSeek V3.1](/models/deepseek-v3-1) | DeepSeek | 46.3 | 31.5-61.0 | Supported | Free | 128K | 3 |
| #49 | [Gemini 2.5 Pro](/models/gemini-2-5-pro) | Google | 45.3 | 29.6-61.0 | Supported | $1.25 / $10.00 | 1M | 3 |
| #50 | [Mistral Large 2](/models/mistral-large-2) | Mistral | 44.2 | 26.3-62.1 | Supported | - | 128K | 3 |
| #51 | [Grok 3 Mini](/models/grok-3-mini) | xAI | 43.7 | 30.4-57.0 | Supported | $0.30 / $0.50 | 128K | 3 |
| #52 | [Grok 4 Fast (Reasoning)](/models/grok-4-fast-reasoning) | xAI | 43.4 | 28.4-58.3 | Supported | - | 2M | 2 |
| #53 | [Gemma 4 31B](/models/gemma-4-31b) | Google | 43.0 | 32.0-54.0 | Supported | Free | 256K | 6 |
| #54 | [Gemma 4 26B A4B](/models/gemma-4-26b-a4b) | Google | 40.9 | 30.3-51.5 | Supported | Free | 256K | 6 |
| #55 | [Claude 3.5 Sonnet](/models/claude-3-5-sonnet) | Anthropic | 40.8 | 19.7-61.9 | Supported | $3.00 / $15.00 | 200K | 2 |
| #56 | [Claude 4 Sonnet](/models/claude-4-sonnet) | Anthropic | 40.5 | 19.4-61.7 | Supported | $3.00 / $15.00 | 200K | 2 |
| #57 | [Gemini 3.1 Flash-Lite](/models/gemini-3-1-flash-lite) | Google | 40.3 | 22.8-57.8 | Supported | $0.25 / $1.50 | 1M | 2 |
| #58 | [GPT-5 mini](/models/gpt-5-mini) | OpenAI | 40.3 | 17.9-62.6 | Supported | $0.25 / $2.00 | 128K | 2 |
| #59 | [GPT-OSS 120B](/models/gpt-oss-120b) | OpenAI | 38.3 | 26.9-49.6 | Supported | Free | 128K | 4 |
| #60 | [Gemma 3 27B](/models/gemma-3-27b) | Google | 37.4 | 22.9-51.9 | Supported | Free | 32K | 3 |
| #61 | [GPT-4.1](/models/gpt-4-1) | OpenAI | 37.0 | 22.3-51.6 | Supported | $2.00 / $8.00 | 1M | 3 |
| #62 | [GPT-4o mini](/models/gpt-4o-mini) | OpenAI | 33.6 | 19.6-47.7 | Supported | $0.15 / $0.60 | 128K | 3 |
| #63 | [Kimi K2](/models/kimi-k2) | Moonshot AI | 33.3 | 17.1-49.6 | Estimated | $0.60 / $2.50 | 128K | 2 |
| #64 | [GPT-OSS 20B](/models/gpt-oss-20b) | OpenAI | 32.8 | 20.0-45.7 | Supported | Free | 128K | 3 |
| #65 | [GPT-4o](/models/gpt-4o) | OpenAI | 32.2 | 12.4-52.0 | Supported | $2.50 / $10.00 | 128K | 3 |

## Scored domain benchmarks (10)

- [BullshitBench v2](/benchmarks/bullshitbenchv2) (16% ref. weight): A benchmark that tests whether AI models challenge nonsensical, ill-posed, or logically flawed prompts instead of confidently generating incorrect answers. Measures the critical ability to push back on bad input.
- [FACTS Grounding v2](/benchmarks/facts-grounding) (16% ref. weight): FACTS Grounding v2 checks whether long-form answers remain supported by supplied documents. Its revised judge ensemble and public/private results remain separate from version 1.
- [SimpleQA Verified (Kaggle)](/benchmarks/simpleqa-verified) (16% ref. weight): Kaggle's continuously updated SimpleQA Verified board, with 56 configurations against the 13 rows transcribed from the author paper.
- [FACTS Benchmark Suite](/benchmarks/facts) (15% ref. weight): The official FACTS suite combines document grounding, visual factuality, parametric recall, and search-assisted answering. The published average stays separate from its component scores.
- [FACTS Parametric (Kaggle board)](/benchmarks/factsparametric) (14% ref. weight): FACTS Parametric asks hard factoid questions from real Google searches and checks whether a model answers them from memory, without tools. It is the parametric-knowledge part of the FACTS suite.
- [SimpleQA (Kaggle)](/benchmarks/simpleqa) (13% ref. weight): Kaggle's run of OpenAI's original SimpleQA, separate from provider-reported SimpleQA rows and from the Verified subset.
- [HLE-Verified](/benchmarks/hleverified) (5% ref. weight): A verified and revised Humanity's Last Exam set that removes uncertain items and repairs fixable questions before evaluating frontier expert reasoning.
- [FACTS Grounding V1](/benchmarks/facts-grounding) (2% ref. weight): The archived FACTS Grounding V1 protocol tests factual responses grounded in supplied documents. Its original judge setup and results remain separate from Grounding V2.
- [SimpleQA Verified](/benchmarks/simpleqa-verified) (2% ref. weight): SimpleQA Verified measures short-form factual recall on 1,000 prompts without tools. This dated author-paper table ranks source models by F1 and keeps accuracy, accuracy given attempted, attempted, and hedged separate.
- [SimpleQA](/benchmarks/simpleqa) (1% ref. weight): A benchmark that evaluates the ability of language models to answer short, fact-seeking questions accurately. Focuses on factual correctness rather than reasoning complexity.

## Grounded faithfulness, parametric precision, and pushback against false premises

Ten protocols drive this ranking, led by BullshitBench v2 (16%), FACTS Grounding v2 (16%), SimpleQA Verified (Kaggle) (16%) together with the full FACTS Benchmark Suite, FACTS Parametric, and HLE-Verified. Unlike broad knowledge exams that reward confident guessing on multiple-choice items, these suites penalize unforced fabrications: SimpleQA Verified grades short-form factual precision, FACTS Grounding checks whether every claim stays inside the supplied passage, and BullshitBench v2 tests whether a model challenges nonsensical premises instead of playing along.

Sixty-three of the 65 ranked models carry Supported status here because independent Kaggle and benchmark-owner evaluations cover almost the entire frontier cohort on identical prompts.

## Low benchmark hallucination does not fix stale cutoffs or broken retrieval

Even Gemini 4 Carbon at 78.8 still misses roughly one in five frontier factuality stress items on raw benchmark scales. Parametric recall freezes at each model's training cutoff, and grounded faithfulness benchmarks only verify consistency with the text you paste into the prompt—if your retriever fetches an outdated policy page, a faithful model will repeat that outdated rule accurately.

Treat this leaderboard as a measure of calibration discipline and instruction adherence under uncertainty, not as a substitute for citation verification, live search, or deterministic database lookups on regulated facts.

[Low-hallucination selection guide](/best/factuality) · [Gemini 4 Carbon vs GPT-6 Astra](/compare/gemini-4-carbon-vs-gpt-6-astra)

## Picking a factual model for RAG pipelines versus closed-book chat

Gemini 4 Carbon (78.8, Supported, $2/$10 per 1M tokens) and GPT-6 Astra (77.9, Supported, $10/$50 per 1M tokens) sit 0.9 points apart at the top. If your application supplies retrieved passages and runs millions of daily checks, Gemini 3.8 Flash (#6, 72.6 at $0.75/$3.75 per 1M tokens) preserves over 90% of the leader's calibrated score at a fraction of the API bill.

Open-weight buyers should note the wider spread on closed-book recall: DeepSeek V4.1 Flash 1017 leads open weights at #18 (61.9), because smaller parameter counts store fewer long-tail entity facts even when their grounded RAG discipline is solid.

[Search & Deep Research leaderboard](/leaderboards/search) · [Best open-weight LLMs](/best/open-source)

## Questions

### Which LLM is best for factuality right now?

Gemini 4 Carbon holds #1 on our factuality leaderboard at 78.8 (Supported) as of November 21, 2026, ahead of GPT-6 Astra at 77.9 and Gemini 4 Argon at 77.1. Both top models hold Supported status across independent public evaluations. Check the ranked table and benchmark weights on this page to verify source coverage before production deployment.

### What is the best open-weight LLM for factuality?

DeepSeek V4.1 Flash 1017 from DeepSeek leads open-weight models on this factuality board at #18 with a calibrated score of 61.9 (Supported). That trails Gemini 4 Carbon by 16.9 points, so choose DeepSeek V4.1 Flash 1017 when self-hosting, VPC data residency, or weight customization matters more than peak frontier accuracy.

### Which API model offers the best value for factuality?

Gemini 3.8 Flash (#6, 72.6) is our top API value pick for factuality, priced at $0.75/$3.75 per 1M tokens compared with Gemini 4 Carbon at $2/$10 per 1M tokens. You give up 6.2 calibrated points, making Gemini 3.8 Flash the better buy whenever those extra points do not cut downstream retries.

### Which benchmarks make up the factuality leaderboard score?

BenchAlign v6.0 calibrates 10 open and primary domain protocols for this board, led by BullshitBench v2 (16%), FACTS Grounding v2 (16%), SimpleQA Verified (Kaggle) (16%), FACTS Benchmark Suite (15%). Each percentage reflects share of the vertical's reference weight inside our item-response model rather than a raw score average, and no single independent evaluator can exceed 40% when three

### What is the difference between parametric recall and grounded factuality?

Parametric recall tests whether a model answers closed-book questions accurately from memory without guessing, as measured by SimpleQA Verified and FACTS Parametric. Grounded factuality tests whether answers stay strictly inside supplied source documents without inventing unsupported claims, measured by FACTS Grounding and BullshitBench v2.

### Does a high factuality score eliminate hallucinations in production?

No. Even Gemini 4 Carbon at 78.8 still produces errors on adversarial closed-book prompts and cannot know events after its training cutoff without retrieval. Pair top-ranked models with verified source passages, require inline citations, and abstain or route to human review when retrieved context is missing.

## Related benchmarks and guides

- [Best LLMs for Factuality](/best/factuality): Task guide for low-hallucination chat, citation grounding, and abstention.
- [Knowledge Leaderboard](/knowledge): Academic recall and expert subject-matter rankings.
- [Search & Deep Research Leaderboard](/leaderboards/search): Models ranked on live web browsing and multi-hop citation synthesis.
- [Benchmark Confidence](/benchmark-confidence): Inspect source provenance and verification coverage across ranked models.

## Other specialist vertical leaderboards

- [Cybersecurity leaderboard](/leaderboards/cybersecurity) — #1 Gemini 4 Carbon (90.0)
- [Documents and extraction leaderboard](/leaderboards/documents) — #1 GPT-6.2 Astra (86.5)
- [Finance and accounting leaderboard](/leaderboards/finance) — #1 Gemini 4 Carbon (89.2)
- [Frontend and web development leaderboard](/leaderboards/frontend) — #1 Claude Fable 5.2 (89.5)
- [Healthcare and medical leaderboard](/leaderboards/healthcare) — #1 Claude Fable 5.2 (87.5)
- [Legal leaderboard](/leaderboards/legal) — #1 Claude Fable 5.2 (86.1)
- [Science and research leaderboard](/leaderboards/science) — #1 Claude Fable 5.2 (94.1)
- [Search and deep research leaderboard](/leaderboards/search) — #1 Claude Fable 5.2 (86.6)

Last updated: November 21, 2026

Canonical page: https://benchlm.ai/leaderboards/factuality
