Claude Fable 5
- LMArena rank
- #1 · 1506.8
- Agent Arena
- #1 · Claude Fable 5 (High)
- OpenRouter usage
- #24 · 449.9B tokens/week
- OpenRouter price
- $10.00 in / $50.00 out per 1M
MODEL GAUNTLET RESEARCH DESK
CURRENT RANKING SNAPSHOTUPDATED JULY 16, 2026
SOURCE-ATTRIBUTED AI RANKING AGGREGATOR EDITORIALLY INDEPENDENT
Who’s winningthe AI race?
Anthropic leads the July 16, 2026 LMArena text ranking.
The current frontier
The models in the race right now.
An editorial watchlist of the models in active frontier discussion, audited July 22, 2026: every pick appears on at least one current reviewed board. Selection is ours; every number on a card is one named source's own value with its own scale, every card carries the same stat rows, and a board that does not list a model says so. Nothing is blended.
LMArena rating · style-controlled text · published July 20, 2026
Bars start at zero on LMArena's own scale, so close ratings look close. Hover or focus a bar for the reported interval and battle count. The cards below carry the same values as text.
Claude Mythos 5
- LMArena rank
- Not on this board
- Agent Arena
- Not on this board
- OpenRouter usage
- No matched listing
- OpenRouter price
- Not listed
Claude Opus 4.8 Thinking
- LMArena rank
- #12 · 1484.6
- Agent Arena
- #2 · Claude Opus 4.8 (Thinking)
- OpenRouter usage
- No matched listing
- OpenRouter price
- Not listed
Claude Sonnet 5 High
- LMArena rank
- #37 · 1461.4
- Agent Arena
- #5 · Claude Sonnet 5 (High)
- OpenRouter usage
- No matched listing
- OpenRouter price
- Not listed
GPT-5.6 Sol X-High
- LMArena rank
- #9 · 1486.4
- Agent Arena
- #3 · GPT 5.6 Sol (xHigh)
- OpenRouter usage
- No matched listing
- OpenRouter price
- Not listed
GPT-5.5 High
- LMArena rank
- #13 · 1481.0
- Agent Arena
- #9 · GPT 5.5 (High)
- OpenRouter usage
- No matched listing
- OpenRouter price
- Not listed
Gemini 3.5 Flash High
- LMArena rank
- #15 · 1476.3
- Agent Arena
- #21 · Gemini 3.5 Flash (High)
- OpenRouter usage
- No matched listing
- OpenRouter price
- Not listed
Gemini 3.1 Pro Preview
- LMArena rank
- #11 · 1485.6
- Agent Arena
- #20 · Gemini 3.1 Pro Preview
- OpenRouter usage
- #37 · 174.6B tokens/week
- OpenRouter price
- $2.00 in / $12.00 out per 1M
- EQ-Bench 3
- #19 · 1302
7 Epoch hub benchmarks carry this model. See the benchmark section.
Grok 4.5
- LMArena rank
- #35 · 1465.7
- Agent Arena
- #13 · Grok 4.5
- OpenRouter usage
- #28 · 355.1B tokens/week
- OpenRouter price
- $2.00 in / $6.00 out per 1M
DeepSeek V4 Pro
- LMArena rank
- #43 · 1457.1
- Agent Arena
- #24 · DeepSeek V4 Pro
- OpenRouter usage
- #6 · 2.8T tokens/week
- OpenRouter price
- $0.43 in / $0.87 out per 1M
- EQ-Bench 3
- #5 · 1330
GLM 5.2 (Max)
- LMArena rank
- #29 · 1469.5
- Agent Arena
- #11 · GLM 5.2 (Max)
- OpenRouter usage
- No matched listing
- OpenRouter price
- Not listed
Kimi K3
- LMArena rank
- Not on this board
- Agent Arena
- #4 · Kimi K3
- OpenRouter usage
- No matched listing
- OpenRouter price
- Not listed
Qwen3.7 Max Preview
- LMArena rank
- #18 · 1475.3
- Agent Arena
- Not on this board
- OpenRouter usage
- No matched listing
- OpenRouter price
- Not listed
MiMo V2.5
- LMArena rank
- #78 · 1433.1
- Agent Arena
- Not on this board
- OpenRouter usage
- #2 · 9.4T tokens/week
- OpenRouter price
- Not listed
Sources: LMArena category artifact published July 20, 2026, Aider polyglot updated May 22, 2026, EQ-Bench 3 updated May 10, 2026, and the Epoch AI Benchmarking Hub retrieved July 19, 2026. Each carries its receipt in its labeled section below. Rows named with a scaffold are system results, not bare-model results.
Chatbot Arena archive
Sampled provider frontier ratings, May 2023 to August 2025
Meta highlighted. 827 to 1288 published rating; sampled peak 1396.
Line ends, top to bottom:Google 1466Mistral 1433xAI 1431OpenAI 1429DeepSeek 1424Anthropic 1417Meta 1288
Anthropic Leads LMArena’s July 2026 Text Style-Control Snapshot at 1507.5
Anthropic's Claude Fable 5 holds the highest provider-frontier point estimate in LMArena's July 16, 2026 text style-control overall snapshot, with Meta's Muse Spark 1.1 in second place.
· Leaderboard update
Anthropic's Claude Fable 5 scores 1507.5 in LMArena's July 16, 2026 snapshot
· Model release
OpenAI releases GPT-5.5 on April 23, 2026
· Model release
Google releases Gemini Flash 3.1 TTS on April 15, 2026
· Model release
Meta releases Muse Spark on April 8, 2026
· Model release
Current LMArena snapshot
See who leads this source, and how close the race really is.
These values come from LMArena’s text style-control overall table published July 16, 2026. The legacy archive remains separate.
Anthropic leads
Claude Fable 5 is the highest-scoring model in LMArena’s July 16, 2026 snapshot, with a reported interval of 1500.2 to 1514.7.
Gap to second
Meta’s Muse Spark 1.1 ranks second at 1493.1, 14.4 points behind Anthropic.
Leader battles
LMArena reports 8,817 battles for Claude Fable 5 in this published snapshot.
Tracked organizations
Each model is matched to the organization listed in LMArena’s published data, avoiding guesses based on model names.
Point-in-time standings
The leading model from each AI lab in this source.
We take each organization’s highest LMArena rating in this published snapshot. Confidence intervals show where close results may not represent a meaningful difference.
| Source rank | Model | Organization | LMArena rating | Reported interval | Battles |
|---|---|---|---|---|---|
| 01 | Claude Fable 5 |  Anthropic | 1507.5 | 1500.2–1514.7 | 8,817 |
| 02 | Muse Spark 1.1 |  Meta | 1493.1 | 1484.9–1501.3 | 5,732 |
| 03 | GPT-5.6 Sol X-High |  OpenAI | 1486.3 | 1476.8–1495.8 | 4,113 |
| 04 | Gemini 3 Pro |  Google | 1485.8 | 1481.9–1489.6 | 41,283 |
| 05 | Grok 4.20 Beta 1 |  xAI | 1474.2 | 1469.5–1478.8 | 26,844 |
| 06 | DeepSeek V4 Pro |  DeepSeek | 1457.5 | 1453.1–1462.0 | 43,062 |
| 07 | Mistral Medium 3.5 |  Mistral | 1427.3 | 1420.7–1433.9 | 11,017 |
LMArena category boards
Six task boards from the same named source.
Each board is LMArena's own category rating. Open one board at a time; every board links to its full page with intervals and history.
Coding arena
Creative writing arena
Instruction following arena
Hard prompts arena
Multi-turn arena
Long queries arena
Math arena
Vision arena
Web development arena
Search arena
Document arena
Aider polyglot · code editing
Code editing systems, ranked by their own leaderboard.
Aider ranks full model-and-editor system configurations on code editing, not bare models. Pass rate 2 is Aider's own metric, and these rows are never compared against LMArena ratings.
Agent Arena · agentic tasks
The agent board, scored by its own arena.
LMArena's Agent Arena ranks models on real agentic sessions and published this board July 20, 2026.IPS is this arena's own score with its own scale. Rows labeled with a reasoning configuration are system results for that configuration, and a score at or below zero draws no bar.
EQ-Bench 3 · emotional intelligence
One source's judge-scored emotional intelligence board.
EQ-Bench 3 publishes its own Elo from judged role-play scenarios. The scale belongs to this source alone and shares nothing with LMArena numbers.
Epoch AI Benchmarking Hub
Ten benchmarks, reported one at a time.
Every card is one benchmark's own top scores from Epoch AI's published hub data. Rows labeled with a scaffold, such as a coding agent or edit format, are system results, not bare-model results. Fraction-unit scores are shown as percentages.
Aider Polyglot
Aider Polyglot leaderboard · 69 resolved scores
Chess Puzzles
Epoch AI Benchmarking Hub · 54 resolved scores
FrontierMath
Epoch AI Benchmarking Hub · 98 resolved scores
FrontierMath Tier 4
Epoch AI Benchmarking Hub · 69 resolved scores
GPQA Diamond
Epoch AI Benchmarking Hub · 176 resolved scores
MATH Level 5
Epoch AI Benchmarking Hub · 106 resolved scores
OTIS Mock AIME 2024-2025
Epoch AI Benchmarking Hub · 150 resolved scores
SimpleQA Verified
Epoch AI Benchmarking Hub · 62 resolved scores
SWE-bench Verified
Epoch AI Benchmarking Hub · 33 resolved scores
Terminal-Bench
Terminal-Bench leaderboard · 178 resolved scores
OpenRouter platform view
What the platform actually routes, lists, and charges.
OpenRouter's own usage rankings, model listings, and routed prices. Usage is the platform's weekly routed token volume,retrieved July 22, 2026: a live popularity signal that shows which models developers route work through right now, not a quality score. The platform's embedded third-party benchmark data is excluded at collection with a receipt and never appears on this site.
Most-used models this week
OpenRouter's routed token volume, week view. Volume is not a quality signal.
Newest listings
By OpenRouter's listing date. A listing date is not a release date.
- Poolside: Laguna S 2.12026-07-21
- Google: Gemini 3.6 Flash2026-07-21
- Google: Gemini 3.5 Flash-Lite2026-07-21
- Meituan: LongCat 2.02026-07-20
- Thinking Machines: Inkling2026-07-17
- Auto Router (Beta)2026-07-17
- MoonshotAI: Kimi K32026-07-16
- Meta: Muse Spark 1.12026-07-16
- Kwaipilot: KAT-Coder-Air V2.52026-07-10
- Kwaipilot: KAT-Coder-Pro V2.52026-07-10
Frontier watchlist, priced
Input price per million tokens, cheapest first. Price says nothing about quality.
From the archive
How Meta moved through the historical AI race.
827 → 1396 → 1288
Meta’s reviewed historical path
This history includes Meta’s own models and excludes third-party models built from Llama.
Chart lab
Compare leads, winning streaks, and head-to-head movement
Explore the race charts →Methodology
What is measured, and what is not
Read the methodology →Explore Model Gauntlet
The race desk.
Rankings, model records, research reports, and interactive history. Every claim is linked to its source, with clear evidence boundaries.
Reports · 4 publishedSource-backed analysis you can read, share, and cite.Open the research desk →Models · 456 sourced recordsSearch releases, labs, access, tasks, and evidence.Explore models →Interactive race reportReplay major shifts, one dated snapshot at a time.Enter race replay →Company race · 5 evidence filesCompare the financial scale behind major AI companies using SEC filings.Inspect company evidence →
Research and commercial inquiries
Work with the team behind the research.
Research questions, source submissions, corrections, or partnerships. Commercial relationships never influence a score, ranking, or editorial conclusion.
Direct answers
How Model Gauntlet works.
What is Model Gauntlet?
Model Gauntlet publishes sourced AI model standings and a separately labeled historical race archive. Current and legacy metrics are never blended.
How current are the standings?
The current provider table uses LMArena's published text style-control overall snapshot, dated July 16, 2026. It includes confidence intervals and battle counts.
Does the current table continue the historical chart?
No. The current LMArena rating and the legacy May 2023 to August 2025 archive use incompatible methodology and inclusion rules. No cross-window change is calculated.
Race alerts
Get the important moves in the AI race.
Join the waitlist for sourced updates on leaderboard changes, major model releases, and new research. The briefing launches after a final editorial review; early subscribers get the first edition.