# Model Gauntlet: full machine-readable companion

> This file expands https://modelgauntlet.com/llms.txt with the complete current leaderboard table and key facts for every published model, lab, guide, report, briefing, and news page. Every figure is quoted from a registered source artifact and carries its publication date. Missing measurements stay missing; nothing here is estimated.

Canonical site: https://modelgauntlet.com/
Compact index: https://modelgauntlet.com/llms.txt

## Current leaderboard: all tracked organizations

According to LMArena's published snapshot (text style-control subset, overall category, published 2026-07-12), the leading model per tracked organization is:

| Rank | Organization | Leading model | LMArena rating | Reported 95% CI | Battles |
|---|---|---|---|---|---|
| 1 | Anthropic | claude-fable-5 | 1507.5 | 1500.0 to 1515.0 | 7,856 |
| 2 | Meta | muse-spark-1.1 | 1491.6 | 1482.5 to 1500.7 | 4,498 |
| 3 | Google | gemini-3-pro | 1485.9 | 1482.0 to 1489.7 | 41,290 |
| 4 | OpenAI | gpt-5.6-sol-xhigh | 1484.6 | 1473.2 to 1496.1 | 2,821 |
| 5 | xAI | grok-4.20-beta1 | 1474.4 | 1469.7 to 1479.0 | 26,877 |
| 6 | DeepSeek | deepseek-v4-pro | 1457.1 | 1452.6 to 1461.6 | 41,800 |
| 7 | Mistral | mistral-medium-3.5 | 1427.2 | 1420.6 to 1433.8 | 11,034 |

Attribution: lmarena-ai/leaderboard-dataset (https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset), revision 4a6c22ada142f24563a2ecc5ceb1ce94ab8be019, license CC-BY-4.0. This point-in-time style-control snapshot is not continuous with the May 2023 to August 2025 legacy archive. Do not calculate or visualize cross-window changes.
This is a single-source point-in-time view, not a Model Gauntlet consensus score.
Canonical page: https://modelgauntlet.com/leaderboard/

## Published model profiles (20)

Facts below come from the reviewed Epoch AI catalog artifact (Epoch AI Data on AI Models, CC-BY, retrieved 2026-07-14). They describe models; they do not rank them.

### GPT-5.5
- Canonical URL: https://modelgauntlet.com/models/gpt-5-5/
- Developer: OpenAI
- Released: 2026-04-23 (according to the reviewed Epoch AI catalog artifact sha256:3c663b17ec989f00ef5f994214b878a914118e2d1fb6077e23329a318209eb86)
- Domains: Multimodal, Language, Vision
- Tasks: Language modeling/generation, Question answering
- Access: API access
- Weights: Closed
- Summary: A closed-weight OpenAI API model documented for language, vision, and multimodal work.

### GPT Image 2
- Canonical URL: https://modelgauntlet.com/models/gpt-image-2/
- Developer: OpenAI
- Released: 2026-04-21 (according to the reviewed Epoch AI catalog artifact sha256:3c663b17ec989f00ef5f994214b878a914118e2d1fb6077e23329a318209eb86)
- Domains: Image generation
- Tasks: Image generation
- Access: API access
- Weights: Closed
- Summary: A closed-weight OpenAI image-generation model available through API access.

### Claude Fable 5
- Canonical URL: https://modelgauntlet.com/models/claude-fable-5/
- Developer: Anthropic
- Released: 2026-06-09 (according to the reviewed Epoch AI catalog artifact sha256:3c663b17ec989f00ef5f994214b878a914118e2d1fb6077e23329a318209eb86)
- Domains: Language, Multimodal
- Tasks: Language modeling/generation, Question answering
- Access: API access
- Weights: Closed
- Summary: A closed-weight Anthropic API model documented for language and multimodal question-answering work.
- LMArena observation: a separate pinned LMArena snapshot records 1507.5 for the exact source model ID. This is a point-in-time arena observation, not a composite score.

### Gemini 3.1 Pro
- Canonical URL: https://modelgauntlet.com/models/gemini-3-1-pro/
- Developer: Google
- Released: 2026-02-19 (according to the reviewed Epoch AI catalog artifact sha256:3c663b17ec989f00ef5f994214b878a914118e2d1fb6077e23329a318209eb86)
- Domains: Language, Vision
- Tasks: Language modeling/generation
- Access: API access
- Weights: Closed
- Summary: A closed-weight Google API model documented across language and vision domains.

### Gemini Flash 3.1 TTS
- Canonical URL: https://modelgauntlet.com/models/gemini-flash-3-1-tts/
- Developer: Google
- Released: 2026-04-15 (according to the reviewed Epoch AI catalog artifact sha256:3c663b17ec989f00ef5f994214b878a914118e2d1fb6077e23329a318209eb86)
- Domains: Audio
- Tasks: Audio generation
- Access: API access
- Weights: Closed
- Summary: A closed-weight Google audio-generation model available through API access.

### Llama 4 Scout
- Canonical URL: https://modelgauntlet.com/models/llama-4-scout/
- Developer: Meta
- Released: 2025-04-05 (according to the reviewed Epoch AI catalog artifact sha256:3c663b17ec989f00ef5f994214b878a914118e2d1fb6077e23329a318209eb86)
- Domains: Multimodal, Language, Vision
- Tasks: Chat, Code generation, Visual question answering, Language modeling/generation, Question answering
- Access: Open weights (restricted use)
- Weights: Open
- Summary: A restricted-use open-weight Meta model documented for multimodal, language, vision, chat, code, and question-answering tasks.

### DeepSeek-V4-Pro
- Canonical URL: https://modelgauntlet.com/models/deepseek-v4-pro/
- Developer: DeepSeek
- Released: 2026-04-24 (according to the reviewed Epoch AI catalog artifact sha256:3c663b17ec989f00ef5f994214b878a914118e2d1fb6077e23329a318209eb86)
- Domains: Language
- Tasks: Language modeling/generation, Question answering
- Access: Open weights (unrestricted)
- Weights: Open
- Summary: An unrestricted open-weight DeepSeek language model documented for generation and question answering.
- LMArena observation: a separate pinned LMArena snapshot records 1457.1 for the exact source model ID. This is a point-in-time arena observation, not a composite score.

### Grok 4.1
- Canonical URL: https://modelgauntlet.com/models/grok-4-1/
- Developer: xAI
- Released: 2025-11-17 (according to the reviewed Epoch AI catalog artifact sha256:3c663b17ec989f00ef5f994214b878a914118e2d1fb6077e23329a318209eb86)
- Domains: Language
- Tasks: Language modeling/generation, Question answering, Mathematical reasoning, Quantitative reasoning, Code generation
- Access: API access
- Weights: Closed
- Summary: A closed-weight xAI API model documented for language, question answering, math, quantitative reasoning, and code generation.

### Mistral OCR
- Canonical URL: https://modelgauntlet.com/models/mistral-ocr/
- Developer: Mistral
- Released: 2025-03-06 (according to the reviewed Epoch AI catalog artifact sha256:3c663b17ec989f00ef5f994214b878a914118e2d1fb6077e23329a318209eb86)
- Domains: Multimodal, Vision, Language
- Tasks: Character recognition (OCR), Chat, Language generation
- Access: API access
- Weights: Closed
- Summary: A closed-weight Mistral API model documented for OCR, chat, and language generation across vision, language, and multimodal domains.

### Pixtral Large
- Canonical URL: https://modelgauntlet.com/models/pixtral-large/
- Developer: Mistral
- Released: 2024-11-18 (according to the reviewed Epoch AI catalog artifact sha256:3c663b17ec989f00ef5f994214b878a914118e2d1fb6077e23329a318209eb86)
- Domains: Multimodal, Language, Vision
- Tasks: Vision-language generation, Visual question answering, Mathematical reasoning, Character recognition (OCR), Language modeling/generation, Question answering
- Access: Open weights (non-commercial)
- Weights: Open
- Summary: A non-commercial open-weight Mistral model documented for multimodal language and vision work.

### GPT-5.3 Codex
- Canonical URL: https://modelgauntlet.com/models/gpt-5-3-codex/
- Developer: OpenAI
- Released: 2026-02-05 (according to the reviewed Epoch AI catalog artifact sha256:3c663b17ec989f00ef5f994214b878a914118e2d1fb6077e23329a318209eb86)
- Domains: Language
- Tasks: Language modeling/generation
- Access: API access
- Weights: Closed
- Summary: A closed-weight OpenAI API model in the Codex line, documented for language modeling and generation.

### Claude Opus 4.7
- Canonical URL: https://modelgauntlet.com/models/claude-opus-4-7/
- Developer: Anthropic
- Released: 2026-04-16 (according to the reviewed Epoch AI catalog artifact sha256:3c663b17ec989f00ef5f994214b878a914118e2d1fb6077e23329a318209eb86)
- Domains: Language
- Tasks: Question answering, Language modeling/generation
- Access: API access
- Weights: Closed
- Summary: A closed-weight Anthropic API model documented for question answering and language generation work.

### Claude Haiku 4.5
- Canonical URL: https://modelgauntlet.com/models/claude-haiku-4-5/
- Developer: Anthropic
- Released: 2025-10-15 (according to the reviewed Epoch AI catalog artifact sha256:3c663b17ec989f00ef5f994214b878a914118e2d1fb6077e23329a318209eb86)
- Domains: Language
- Tasks: Chat, Code generation, Language modeling/generation, Question answering
- Access: API access
- Weights: Closed
- Summary: A closed-weight Anthropic API model documented for chat, code generation, language generation, and question answering.

### Gemini 3 Pro Image (Nano Banana Pro)
- Canonical URL: https://modelgauntlet.com/models/gemini-3-pro-image/
- Developer: Google
- Released: 2025-11-20 (according to the reviewed Epoch AI catalog artifact sha256:3c663b17ec989f00ef5f994214b878a914118e2d1fb6077e23329a318209eb86)
- Domains: Image generation
- Tasks: Image generation
- Access: API access
- Weights: Closed
- Summary: A closed-weight Google image-generation model, also named Nano Banana Pro, available through API access.

### Muse Spark
- Canonical URL: https://modelgauntlet.com/models/muse-spark/
- Developer: Meta
- Released: 2026-04-08 (according to the reviewed Epoch AI catalog artifact sha256:3c663b17ec989f00ef5f994214b878a914118e2d1fb6077e23329a318209eb86)
- Domains: Multimodal, Language, Vision
- Tasks: Language modeling/generation, Question answering
- Access: API access
- Weights: Closed
- Summary: A closed-weight Meta API model documented across multimodal, language, and vision domains for generation and question answering.

### Llama 4 Maverick
- Canonical URL: https://modelgauntlet.com/models/llama-4-maverick/
- Developer: Meta
- Released: 2025-04-05 (according to the reviewed Epoch AI catalog artifact sha256:3c663b17ec989f00ef5f994214b878a914118e2d1fb6077e23329a318209eb86)
- Domains: Multimodal, Language, Vision
- Tasks: Chat, Code generation, Visual question answering, Language modeling/generation, Question answering
- Access: Open weights (restricted use)
- Weights: Open
- Summary: A restricted-use open-weight Meta model documented for chat, code, visual question answering, and language work across multimodal domains.

### DeepSeek-R1 (May 2025)
- Canonical URL: https://modelgauntlet.com/models/deepseek-r1-may-2025/
- Developer: DeepSeek
- Released: 2025-05-28 (according to the reviewed Epoch AI catalog artifact sha256:3c663b17ec989f00ef5f994214b878a914118e2d1fb6077e23329a318209eb86)
- Domains: Language
- Tasks: Language modeling/generation, Code generation, Quantitative reasoning, Question answering
- Access: Open weights (unrestricted)
- Weights: Open
- Summary: An unrestricted open-weight DeepSeek reasoning-line model documented for language generation, code generation, quantitative reasoning, and question answering.

### DeepSeekMath-V2
- Canonical URL: https://modelgauntlet.com/models/deepseekmath-v2/
- Developer: DeepSeek
- Released: 2025-11-27 (according to the reviewed Epoch AI catalog artifact sha256:3c663b17ec989f00ef5f994214b878a914118e2d1fb6077e23329a318209eb86)
- Domains: Language
- Tasks: Mathematical reasoning, Theorem proving, Proof generation, Proof verification
- Access: Open weights (unrestricted)
- Weights: Open
- Summary: An unrestricted open-weight DeepSeek model documented for mathematical reasoning, theorem proving, and proof generation and verification.

### Grok 4.20
- Canonical URL: https://modelgauntlet.com/models/grok-4-20/
- Developer: xAI
- Released: 2026-02-17 (according to the reviewed Epoch AI catalog artifact sha256:3c663b17ec989f00ef5f994214b878a914118e2d1fb6077e23329a318209eb86)
- Domains: Language
- Tasks: Language modeling/generation
- Access: API access
- Weights: Closed
- Summary: A closed-weight xAI API model documented for language modeling and generation.

### Grok 4
- Canonical URL: https://modelgauntlet.com/models/grok-4/
- Developer: xAI
- Released: 2025-07-09 (according to the reviewed Epoch AI catalog artifact sha256:3c663b17ec989f00ef5f994214b878a914118e2d1fb6077e23329a318209eb86)
- Domains: Language, Multimodal, Vision
- Tasks: Language modeling/generation, Question answering, Search, Visual question answering, Character recognition (OCR), Image captioning, Quantitative reasoning
- Access: API access
- Weights: Closed
- Summary: A closed-weight xAI API model documented across language, multimodal, and vision domains, including search, OCR, and image-captioning tasks.

## Lab files (7)

Current rank and rating come from the LMArena snapshot published 2026-07-12. Legacy observations belong to the separately labeled May 2023 to August 2025 archive and are not comparable to the current metric.

- [Meta](https://modelgauntlet.com/labs/meta/): rank #2 in the current LMArena provider view at 1491.6 (Muse Spark 1.1); 3 reviewed catalog models; 11 separately labeled legacy observations.
- [OpenAI](https://modelgauntlet.com/labs/openai/): rank #4 in the current LMArena provider view at 1484.6 (GPT-5.6 Sol X-High); 3 reviewed catalog models; 11 separately labeled legacy observations.
- [Anthropic](https://modelgauntlet.com/labs/anthropic/): rank #1 in the current LMArena provider view at 1507.5 (Claude Fable 5); 3 reviewed catalog models; 11 separately labeled legacy observations.
- [Google](https://modelgauntlet.com/labs/google/): rank #3 in the current LMArena provider view at 1485.9 (Gemini 3 Pro); 3 reviewed catalog models; 10 separately labeled legacy observations.
- [Mistral](https://modelgauntlet.com/labs/mistral/): rank #7 in the current LMArena provider view at 1427.2 (Mistral Medium 3.5); 2 reviewed catalog models; 9 separately labeled legacy observations.
- [DeepSeek](https://modelgauntlet.com/labs/deepseek/): rank #6 in the current LMArena provider view at 1457.1 (DeepSeek V4 Pro); 3 reviewed catalog models; 7 separately labeled legacy observations.
- [xAI](https://modelgauntlet.com/labs/xai/): rank #5 in the current LMArena provider view at 1474.4 (Grok 4.20 Beta 1); 3 reviewed catalog models; 5 separately labeled legacy observations.

## Canonical head-to-head comparisons (23)

Each comparison page reports documented Epoch AI catalog facts side by side and quotes LMArena ratings only when both models have an exact match in the pinned snapshot. No winner is declared beyond what a named source's numbers say. Hub for the best-model question: [Best AI Model 2026: What the Evidence Actually Supports](https://modelgauntlet.com/best-ai-model/).

- [Claude Fable 5 vs DeepSeek-V4-Pro](https://modelgauntlet.com/vs/claude-fable-5-vs-deepseek-v4-pro/): According to LMArena's snapshot published July 12, 2026, Anthropic's Claude Fable 5 scores 1507.5 and DeepSeek's DeepSeek-V4-Pro scores 1457.1 in the style-controlled text category. Those two numbers are the only shared ranking evidence between these models in Model Gauntlet's approved sources.
- [Claude Fable 5 vs Gemini 3.1 Pro](https://modelgauntlet.com/vs/claude-fable-5-vs-gemini-3-1-pro/): According to LMArena's snapshot published July 12, 2026, Claude Fable 5 scores 1507.5 in the style-controlled text category, but Gemini 3.1 Pro has no exact model match in that snapshot. The two models share no ranking evidence, so this page names no winner.
- [Claude Fable 5 vs GPT-5.5](https://modelgauntlet.com/vs/claude-fable-5-vs-gpt-5-5/): According to LMArena's snapshot published July 12, 2026, Claude Fable 5 scores 1507.5 in the style-controlled text category, but GPT-5.5 has no exact model match in that snapshot. The two models share no ranking evidence, so this page names no winner.
- [Claude Fable 5 vs Grok 4.1](https://modelgauntlet.com/vs/claude-fable-5-vs-grok-4-1/): According to LMArena's snapshot published July 12, 2026, Claude Fable 5 scores 1507.5 in the style-controlled text category, but Grok 4.1 has no exact model match in that snapshot. The two models share no ranking evidence, so this page names no winner.
- [Claude Fable 5 vs Llama 4 Scout](https://modelgauntlet.com/vs/claude-fable-5-vs-llama-4-scout/): According to LMArena's snapshot published July 12, 2026, Claude Fable 5 scores 1507.5 in the style-controlled text category, but Llama 4 Scout has no exact model match in that snapshot. The two models share no ranking evidence, so this page names no winner.
- [Claude Fable 5 vs Pixtral Large](https://modelgauntlet.com/vs/claude-fable-5-vs-pixtral-large/): According to LMArena's snapshot published July 12, 2026, Claude Fable 5 scores 1507.5 in the style-controlled text category, but Pixtral Large has no exact model match in that snapshot. The two models share no ranking evidence, so this page names no winner.
- [DeepSeek-V4-Pro vs Gemini 3.1 Pro](https://modelgauntlet.com/vs/deepseek-v4-pro-vs-gemini-3-1-pro/): According to LMArena's snapshot published July 12, 2026, DeepSeek-V4-Pro scores 1457.1 in the style-controlled text category, but Gemini 3.1 Pro has no exact model match in that snapshot. The two models share no ranking evidence, so this page names no winner.
- [DeepSeek-V4-Pro vs GPT-5.5](https://modelgauntlet.com/vs/deepseek-v4-pro-vs-gpt-5-5/): According to LMArena's snapshot published July 12, 2026, DeepSeek-V4-Pro scores 1457.1 in the style-controlled text category, but GPT-5.5 has no exact model match in that snapshot. The two models share no ranking evidence, so this page names no winner.
- [DeepSeek-V4-Pro vs Grok 4.1](https://modelgauntlet.com/vs/deepseek-v4-pro-vs-grok-4-1/): According to LMArena's snapshot published July 12, 2026, DeepSeek-V4-Pro scores 1457.1 in the style-controlled text category, but Grok 4.1 has no exact model match in that snapshot. The two models share no ranking evidence, so this page names no winner.
- [DeepSeek-V4-Pro vs Llama 4 Scout](https://modelgauntlet.com/vs/deepseek-v4-pro-vs-llama-4-scout/): According to LMArena's snapshot published July 12, 2026, DeepSeek-V4-Pro scores 1457.1 in the style-controlled text category, but Llama 4 Scout has no exact model match in that snapshot. The two models share no ranking evidence, so this page names no winner.
- [DeepSeek-V4-Pro vs Pixtral Large](https://modelgauntlet.com/vs/deepseek-v4-pro-vs-pixtral-large/): According to LMArena's snapshot published July 12, 2026, DeepSeek-V4-Pro scores 1457.1 in the style-controlled text category, but Pixtral Large has no exact model match in that snapshot. The two models share no ranking evidence, so this page names no winner.
- [Gemini 3.1 Pro vs GPT-5.5](https://modelgauntlet.com/vs/gemini-3-1-pro-vs-gpt-5-5/): Neither Gemini 3.1 Pro nor GPT-5.5 appears with an exact model match in the LMArena snapshot published July 12, 2026. The two models share no ranking evidence in Model Gauntlet's approved sources, so this page compares documented Epoch AI catalog facts and names no winner.
- [Gemini 3.1 Pro vs Grok 4.1](https://modelgauntlet.com/vs/gemini-3-1-pro-vs-grok-4-1/): Neither Gemini 3.1 Pro nor Grok 4.1 appears with an exact model match in the LMArena snapshot published July 12, 2026. The two models share no ranking evidence in Model Gauntlet's approved sources, so this page compares documented Epoch AI catalog facts and names no winner.
- [Gemini 3.1 Pro vs Llama 4 Scout](https://modelgauntlet.com/vs/gemini-3-1-pro-vs-llama-4-scout/): Neither Gemini 3.1 Pro nor Llama 4 Scout appears with an exact model match in the LMArena snapshot published July 12, 2026. The two models share no ranking evidence in Model Gauntlet's approved sources, so this page compares documented Epoch AI catalog facts and names no winner.
- [Gemini 3.1 Pro vs Pixtral Large](https://modelgauntlet.com/vs/gemini-3-1-pro-vs-pixtral-large/): Neither Gemini 3.1 Pro nor Pixtral Large appears with an exact model match in the LMArena snapshot published July 12, 2026. The two models share no ranking evidence in Model Gauntlet's approved sources, so this page compares documented Epoch AI catalog facts and names no winner.
- [GPT-5.5 vs Grok 4.1](https://modelgauntlet.com/vs/gpt-5-5-vs-grok-4-1/): Neither GPT-5.5 nor Grok 4.1 appears with an exact model match in the LMArena snapshot published July 12, 2026. The two models share no ranking evidence in Model Gauntlet's approved sources, so this page compares documented Epoch AI catalog facts and names no winner.
- [GPT-5.5 vs Llama 4 Scout](https://modelgauntlet.com/vs/gpt-5-5-vs-llama-4-scout/): Neither GPT-5.5 nor Llama 4 Scout appears with an exact model match in the LMArena snapshot published July 12, 2026. The two models share no ranking evidence in Model Gauntlet's approved sources, so this page compares documented Epoch AI catalog facts and names no winner.
- [GPT-5.5 vs Pixtral Large](https://modelgauntlet.com/vs/gpt-5-5-vs-pixtral-large/): Neither GPT-5.5 nor Pixtral Large appears with an exact model match in the LMArena snapshot published July 12, 2026. The two models share no ranking evidence in Model Gauntlet's approved sources, so this page compares documented Epoch AI catalog facts and names no winner.
- [Grok 4.1 vs Llama 4 Scout](https://modelgauntlet.com/vs/grok-4-1-vs-llama-4-scout/): Neither Grok 4.1 nor Llama 4 Scout appears with an exact model match in the LMArena snapshot published July 12, 2026. The two models share no ranking evidence in Model Gauntlet's approved sources, so this page compares documented Epoch AI catalog facts and names no winner.
- [Grok 4.1 vs Pixtral Large](https://modelgauntlet.com/vs/grok-4-1-vs-pixtral-large/): Neither Grok 4.1 nor Pixtral Large appears with an exact model match in the LMArena snapshot published July 12, 2026. The two models share no ranking evidence in Model Gauntlet's approved sources, so this page compares documented Epoch AI catalog facts and names no winner.
- [Llama 4 Scout vs Mistral OCR](https://modelgauntlet.com/vs/llama-4-scout-vs-mistral-ocr/): Neither Llama 4 Scout nor Mistral OCR appears with an exact model match in the LMArena snapshot published July 12, 2026. The two models share no ranking evidence in Model Gauntlet's approved sources, so this page compares documented Epoch AI catalog facts and names no winner.
- [Llama 4 Scout vs Pixtral Large](https://modelgauntlet.com/vs/llama-4-scout-vs-pixtral-large/): Neither Llama 4 Scout nor Pixtral Large appears with an exact model match in the LMArena snapshot published July 12, 2026. The two models share no ranking evidence in Model Gauntlet's approved sources, so this page compares documented Epoch AI catalog facts and names no winner.
- [Mistral OCR vs Pixtral Large](https://modelgauntlet.com/vs/mistral-ocr-vs-pixtral-large/): Neither Mistral OCR nor Pixtral Large appears with an exact model match in the LMArena snapshot published July 12, 2026. The two models share no ranking evidence in Model Gauntlet's approved sources, so this page compares documented Epoch AI catalog facts and names no winner.

## Purpose guides (5)

Shortlists are membership lists from explicit Epoch task and domain labels (catalog current through 2026-07-14). They are not performance rankings.

- [Which AI models belong on a coding shortlist?](https://modelgauntlet.com/guides/coding/): 5 source-backed models. Five source-backed models whose Epoch records explicitly include code generation. This is a shortlist, not a performance ranking.
- [Which AI models belong on a reasoning or research shortlist?](https://modelgauntlet.com/guides/reasoning-research/): 5 source-backed models. Five source-backed models whose Epoch records explicitly name reasoning, proof, or search work. This is a shortlist, not a performance ranking.
- [Which AI models belong on a writing and marketing shortlist?](https://modelgauntlet.com/guides/writing-marketing/): 16 source-backed models. Source-backed models explicitly labeled for language generation or chat. The labels establish relevance, not writing quality or brand fit.
- [Which AI models belong on a multimodal shortlist?](https://modelgauntlet.com/guides/multimodal/): 12 source-backed models. Source-backed models whose Epoch records explicitly include a multimodal, vision, image, or audio domain. This is a capability map, not a quality ranking.
- [How should you evaluate AI models for cost-sensitive API use?](https://modelgauntlet.com/guides/cost-sensitive-api/): evaluation framework, no shortlist published. A source-bounded evaluation plan for cost-sensitive deployments. No cost shortlist is published because Epoch's catalog does not include pricing.

## Research reports (4)

- [State of the AI Race: Q3 2026](https://modelgauntlet.com/reports/state-of-the-race/2026-q3/): published 2026-07-14, data current through 2026-07-12. Anthropic leads the July 12, 2026 LMArena provider-frontier snapshot. Explore the separate historical archive and the methodology behind both views.
- [Launch Impact: Claude Fable 5 Enters at the Front](https://modelgauntlet.com/reports/launch-impact/claude-fable-5/): published 2026-07-14, data current through 2026-07-12. What the Epoch catalog and July 2026 LMArena snapshot establish about Claude Fable 5, and what they cannot establish about launch impact.
- [Lab Momentum in the Selected Arena Archive](https://modelgauntlet.com/reports/lab-momentum/selected-archive/): published 2026-07-14, data current through 2025-08-29. A bounded reading of provider movement across eleven selected LMArena artifacts from May 2023 through August 2025.
- [Practical Model Selection: Build an Evidence-First Shortlist](https://modelgauntlet.com/reports/practical-model-selection/evidence-first-shortlist/): published 2026-07-14, data current through 2026-07-14. A six-step model-selection workflow using sourced task, domain, access, weights, release, and Arena evidence without a universal score.

## Weekly race briefings (1)

- [Weekly Race Briefing: 2026-07-15](https://modelgauntlet.com/briefings/weekly/2026-07-15/): published 2026-07-15, data current through 2026-07-12. Anthropic leads LMArena by 15.9 points, but the margin is inside the error bars. According to LMArena's July 12, 2026 snapshot, Anthropic holds a narrow lead over a compressed chase group, and the published confidence intervals overlap. Every claim in this briefing links to a named source and its date.

## Reviewed release briefs (5)

- [Claude Fable 5: reviewed release brief](https://modelgauntlet.com/briefings/claude-fable-5/): published 2026-07-14. Anthropic released Claude Fable 5 on June 9, 2026. It now scores 1507.5 in LMArena's July 12, 2026 snapshot, where it is Anthropic's leading model.
- [GPT-5.5: reviewed release brief](https://modelgauntlet.com/briefings/gpt-5-5/): published 2026-07-14. OpenAI released GPT-5.5 on April 23, 2026. It gives OpenAI a new multimodal, language and vision option with API access, according to Epoch AI's catalog.
- [Gemini Flash 3.1 TTS: reviewed release brief](https://modelgauntlet.com/briefings/gemini-flash-3-1-tts/): published 2026-07-14. Google released Gemini Flash 3.1 TTS on April 15, 2026. It gives Google a new audio option with API access, according to Epoch AI's catalog.
- [Muse Spark: reviewed release brief](https://modelgauntlet.com/briefings/muse-spark/): published 2026-07-14. Meta released Muse Spark on April 8, 2026. It gives Meta a new multimodal, language and vision option with API access, according to Epoch AI's catalog.
- [Grok 4.20: reviewed release brief](https://modelgauntlet.com/briefings/grok-4-20/): published 2026-07-14. xAI released Grok 4.20 on February 17, 2026. It gives xAI a new language option with API access, according to Epoch AI's catalog.

## News (1)

The newsroom river at https://modelgauntlet.com/news/ carries these industry stories together with the reviewed release briefs above; each story has exactly one canonical URL.

- [Anthropic Leads the July 12 LMArena Provider View](https://modelgauntlet.com/news/anthropic-leads-july-12-lmarena-provider-view/): published 2026-07-14, data current through 2026-07-12. Anthropic holds the highest provider-frontier point estimate in LMArena's July 12, 2026 text style-control overall snapshot.

## Site updates (1)

Notes about Model Gauntlet itself. They are published below the news river and excluded from the news feeds.

- [Epoch Catalog Powers 20 Reviewed Model Records](https://modelgauntlet.com/news/epoch-catalog-powers-20-reviewed-model-records/): published 2026-07-14, data current through 2026-07-14. Model Gauntlet has published 20 reviewed model records from a pinned Epoch AI catalog artifact with source links and evidence boundaries.

## Data boundaries

- The current point-in-time table uses LMArena text style-control, overall category, published 2026-07-12.
- The historical archive covers 2023-05-08 through 2025-08-29 across 11 selected published snapshots. Methodology changes make cross-window movement non-comparable, so no current-to-legacy change is calculated.
- Why no chart data exists between 2025-08-29 and 2026-07-12: https://modelgauntlet.com/methodology/evidence-gap/
- Model catalog facts come from the reviewed Epoch AI artifact sha256:3c663b17ec989f00ef5f994214b878a914118e2d1fb6077e23329a318209eb86, retrieved 2026-07-14.
- Model Gauntlet does not ingest or republish Artificial Analysis data.
- Corrections policy and log: https://modelgauntlet.com/corrections/