Independent LLM leaderboards change as new models are added and measure different things, so a score quoted without its leaderboard, version and date is hard to check. This snapshot records the published figures of two independent leaderboards, the Artificial Analysis Intelligence Index and the Epoch AI Capabilities Index (ECI), exactly as each leaderboard's own page showed them on 15 November 2026, with each leaderboard's method note. It is written for engineers, buyers and AI governance teams who cite benchmark figures in model selection papers and want a dated, sourced record. Rows are listed alphabetically by model name; the column headed "Leaderboard position" repeats the leaderboard's own order and adds no ranking of ours. Last checked: 2026-11-15.
This is an independent reference page from NHI Mgmt Group based on public provider documentation and the named leaderboards' own pages; it is not sponsored or endorsed by any provider or leaderboard.
Key takeaways
- Two leaderboards, two scales. The Artificial Analysis Intelligence Index v5.0.1 is a weighted average of nine evaluations; Epoch AI's ECI combines more than 50 benchmarks into one capability scale. Their numbers cannot be compared with each other.
- Each figure carries its own date. Epoch AI publishes a date per model row and dated its datasets 14 November 2026; the Artificial Analysis leaderboard page showed no date, so its rows are recorded as checked on 15 November 2026.
- Confidence intervals matter. Epoch AI publishes a confidence interval for every ECI score, and many neighbouring models have overlapping intervals.
- Provider claims are a separate column. Providers often quote leaderboard figures in launch posts; those can predate an index revision, so this page reports only what the leaderboard itself shows.
- Snapshots age fast. Epoch AI's 9 November 2026 update describes the newest top score as based on pre-launch testing that may change as more results become available.
Status key
As of 2026-11-15.
| Term | Meaning on this page |
|---|---|
| Leaderboard position | The row's position in the leaderboard's own published order on the date checked |
| Score | The figure exactly as the leaderboard publishes it, unrounded |
| CI | Confidence interval published by the leaderboard (Epoch AI only) |
| Date | The date the leaderboard attaches to the row; where it shows none, the date checked |
| Not shown | The leaderboard page displays no value in that cell |
Leaderboards covered and their methods
| Leaderboard | Measure | Method note published by the leaderboard | Date shown |
|---|---|---|---|
| Artificial Analysis | Intelligence Index v5.0.1 | Weighted average across four categories at 25% each: knowledge work (AA-Briefcase v1.1, AutomationBench-AA, GDP.pdf), coding (FrontierCode-AA 2.0, Terminal-Bench 4.0), science (CritPt v2 Hard Subset, Terminal-Bench-Science 0.1) and general (AA-Omniscience v1.1, HLE-Diamond). Cost per task uses token counts reported by each model's API provider where available | No date on the leaderboard page; checked 2026-11-15 |
| Epoch AI | Epoch Capabilities Index (ECI) | Combines scores from over 50 benchmarks into one general capability scale, estimating each benchmark's difficulty from models evaluated on several benchmarks, so a model scores higher if it does well on harder benchmarks | Datasets dated 2026-11-14; per-row dates |
Other leaderboards are not reproduced in this snapshot. For what each benchmark inside these indices tests, see How to Read LLM Benchmarks and AI Agent Benchmarks Explained.
Artificial Analysis Intelligence Index snapshot
As of 2026-11-15. The 25 highest rows on the Artificial Analysis models leaderboard, listed alphabetically by the leaderboard's own model name, which includes the reasoning setting tested (for example "Max" or "Xhigh"). Creator is shown as the leaderboard displays it. Cost per task and time per task are the leaderboard's own measures. Source: Artificial Analysis models leaderboard.
| Model (as listed) | Creator | Intelligence Index | Leaderboard position | Cost per task (USD) | Time per task (minutes) | Context window |
|---|---|---|---|---|---|---|
| Claude Fable 5.2 Max, Default Fallback | Anthropic | 65 | 1 | $26.54 | 44.3 | 1M |
| Claude Haiku 5.5 Max | Anthropic | 36 | 13 | $0.41 | 14.8 | 1M |
| Claude Opus 5.5 Max, Default Fallback | Anthropic | 59 | 3 | $10.04 | 25.2 | 1M |
| Claude Sonnet 5.5 Max, Default Fallback | Anthropic | 55 | 6 | $10.02 | 33.6 | 1M |
| DeepSeek V4.1 Flash 1017 Max | DeepSeek | 33 | 18 | $0.60 | 13.9 | 1M |
| Gemini 3.8 Flash High | 31 | 20 | $7.77 | 14.6 | 1M | |
| Gemini 4 Argon High | 50 | 9 | $4.97 | 29.6 | 1M | |
| Gemini 4 Carbon High | 60 | 2 | $5.48 | 31.3 | 1M | |
| Gemini 4 Hydrogen Xhigh | 54 | 7 | $3.16 | 14.7 | 1M | |
| GLM-5.3 Max | Z AI | 32 | 19 | $5.01 | 25.0 | 1M |
| GPT-6 Astra Max | OpenAI | 57 | 4 | $6.24 | 22.8 | 1M |
| GPT-6.1 Sol Max | OpenAI | 55 | 5 | $1.27 | 23.3 | 1M |
| GPT-6.1 Terra Max | OpenAI | 51 | 8 | $0.96 | 14.3 | 1M |
| Grok 4.7 Xhigh | SpaceXAI | 35 | 15 | $7.77 | 26.6 | 500k |
| Inkling 2 Small | Thinking Machines | 31 | 21 | $1.18 | 30.9 | 1M |
| Kimi K3.1 Pro | Kimi | 46 | 11 | $3.68 | 39.7 | 1M |
| Ling 3.1 Flash 1019 | InclusionAI | 30 | 23 | $3.19 | Not shown | 1M |
| MiMo-V2.6-Pro | Xiaomi | 30 | 22 | $0.26 | 35.7 | 1M |
| MiniMax-M3.1-Flash | MiniMax | 29 | 24 | $0.49 | 16.6 | 1M |
| Muse Ember Max | Meta | 36 | 14 | $0.14 | 3.3 | 1M |
| Muse Wonder Full Max | Meta | 50 | 10 | Not shown | Not shown | 1M |
| Muse Wonder Max | Meta | 44 | 12 | $6.95 | 14.3 | 1M |
| Qwen3.8 Max (0902) | Alibaba | 34 | 17 | $10.02 | 48.6 | 984k |
| Qwen3.8-Flash-Next | Alibaba | 29 | 25 | $0.57 | 33.8 | 256k |
| Tencent Hy4 | Tencent | 35 | 16 | $1.03 | 14.2 | 1M |
Ties share a score; where the leaderboard lists tied models, the position column follows the order in which it lists them. Step 5 Preview (StepFun) also scores 29 and is listed directly after the rows above. Several providers have more than one row because Artificial Analysis lists each reasoning setting separately.
Epoch AI Capabilities Index snapshot
As of 2026-11-15 (Epoch AI datasets dated 2026-11-14). The 25 highest ECI scores in Epoch AI's ECI model scores dataset, listed alphabetically by model, with the published confidence interval and the date Epoch AI attaches to each row. Source: Epoch AI Capabilities Index.
| Model | Organisation | ECI | Leaderboard position | CI (low to high) | Date |
|---|---|---|---|---|---|
| Claude Fable 5 | Anthropic | 164.13 | 11 | 161.68 to 166.83 | 2026-06-09 |
| Claude Fable 5.1 | Anthropic | 166.99 | 7 | 164.23 to 170.11 | 2026-09-01 |
| Claude Fable 5.2 | Anthropic | 173.44 | 1 | 169.93 to 177.49 | 2026-11-09 |
| Claude Haiku 5.5 | Anthropic | 160.43 | 18 | 158.27 to 162.87 | 2026-10-07 |
| Claude Opus 4.8 | Anthropic | 159.94 | 21 | 157.86 to 162.2 | 2026-05-28 |
| Claude Opus 5 | Anthropic | 164.68 | 10 | 162.14 to 167.49 | 2026-07-24 |
| Claude Opus 5.5 | Anthropic | 168.54 | 4 | 165.63 to 171.83 | 2026-09-22 |
| Claude Sonnet 5.5 | Anthropic | 166.56 | 8 | 163.76 to 169.58 | 2026-09-28 |
| DeepSeek V4.1 Flash 1017 | DeepSeek | 159.41 | 22 | 157.01 to 162.0 | 2026-10-17 |
| Gemini 3.7 Flash | Google DeepMind | 158.36 | 25 | 156.35 to 160.59 | 2026-08-13 |
| Gemini 3.8 Flash | Google DeepMind | 158.79 | 23 | 156.77 to 160.94 | 2026-09-02 |
| Gemini 4 Argon | Google DeepMind | 166.48 | 9 | 163.77 to 169.46 | 2026-09-30 |
| Gemini 4 Carbon | Google DeepMind | 169.76 | 2 | 166.48 to 173.5 | 2026-10-28 |
| GPT-5.4 Pro | OpenAI | 160.24 | 19 | 157.75 to 163.04 | 2026-03-05 |
| GPT-5.5 | OpenAI | 160.1 | 20 | 158.12 to 162.36 | 2026-04-23 |
| GPT-5.5 Pro | OpenAI | 162.7 | 14 | 159.98 to 165.86 | 2026-04-23 |
| GPT-5.6 Sol | OpenAI | 162.99 | 13 | 160.68 to 165.49 | 2026-07-09 |
| GPT-5.6 Terra | OpenAI | 160.73 | 17 | 158.47 to 163.25 | 2026-07-09 |
| GPT-6 Astra | OpenAI | 169.0 | 3 | 166.08 to 172.26 | 2026-09-03 |
| GPT-6 Sol | OpenAI | 163.75 | 12 | 161.33 to 166.42 | 2026-09-22 |
| GPT-6.1 Sol | OpenAI | 168.48 | 5 | 165.49 to 171.8 | 2026-09-29 |
| GPT-6.1 Terra | OpenAI | 167.67 | 6 | 164.73 to 170.98 | 2026-10-26 |
| Kimi K3 | Moonshot | 158.58 | 24 | 156.6 to 160.8 | 2026-07-16 |
| Kimi K3.1 Pro | Moonshot | 161.37 | 16 | 158.99 to 164.0 | 2026-10-23 |
| Muse Wonder | Meta AI | 161.68 | 15 | 159.36 to 164.34 | 2026-10-12 |
Epoch AI's benchmarks hub, last updated 14 November 2026, states that Claude Fable 5.2 set a new ECI record on 9 November 2026 on pre-launch testing that may change as more results become available. The same hub reports that on FrontierMath Erdős, announced on 3 September 2026, GPT-6.1 Sol had solved 3 of 68 problems (4%). Epoch AI also publishes domain ECIs (for example maths, software engineering and cyber) that reuse the general benchmark parameters; it notes that these domain scores cannot be used alone to judge long-term trends by domain.
Models below the top 25: selected rows
Readers often look up a specific model family that sits outside the top 25. The rows below are the Grok 4.5, 4.6 and 4.7 entries from the same Epoch AI dataset, included because they are compared on a sibling page. Alphabetical.
| Model | Organisation | ECI | CI (low to high) | Date |
|---|---|---|---|---|
| Grok 4.5 | xAI | 154.98 | 153.22 to 156.88 | 2026-07-08 |
| Grok 4.6 | xAI | 157.79 | 155.77 to 159.95 | 2026-08-12 |
| Grok 4.7 | xAI | 156.18 | 154.23 to 158.27 | 2026-09-21 |
See Grok 4.7 vs 4.6 vs 4.5 for the provider's own figures for these versions.
Provider-reported figures versus leaderboard figures
Launch announcements often quote an independent index. Those quotes are dated to the announcement and can refer to an earlier index version. One example from the pages checked:
| Source | Figure | Date | Note |
|---|---|---|---|
| xAI Grok 4.6 announcement | Artificial Analysis Intelligence Index: Grok 4.6 (High) 61, Grok 4.5 (High) 56 | 2026-08-12 | Quoted by the provider in August 2026 |
| Artificial Analysis leaderboard | Intelligence Index v5.0.1: Grok 4.7 Xhigh 35 | Checked 2026-11-15 | Current index version with a different evaluation set |
The two figures sit on different index versions, so the drop in number does not describe a change in the model. When citing any index, record the index version with the score.
What to watch next
- Index revisions: a new Intelligence Index version uses a different evaluation set, so scores from different versions are not comparable.
- Pre-launch results: Epoch AI flags early scores that may change as more results arrive, as with its 9 November 2026 record.
- New releases: models released after 15 November 2026 are not in this snapshot. See the LLM Release Tracker.
- Retirements: leaderboards can keep rows for models a provider has retired. Cross-check with the LLM Deprecation and Retirement Tracker.
How to use this tracker
- Quote the leaderboard, version and date together: for example "Artificial Analysis Intelligence Index v5.0.1, checked 15 November 2026".
- Use confidence intervals: treat Epoch AI models whose intervals overlap as not separable on that measure.
- Match the setting: Artificial Analysis lists reasoning settings separately; compare the setting you will run in production.
- Check the live page before a decision: the figures here are a dated record, not a live feed.
- Run your own evaluation: general indices do not replace tests on your own prompts, data and tools. See Frontier AI Models Compared for specifications and prices.
Practitioner checklist
- Record the leaderboard name, index version, row date and access date beside every score you cite.
- Keep provider-reported benchmark figures in a separate column from independent leaderboard figures.
- Never compare an Artificial Analysis index value with an Epoch AI ECI value.
- Check the reasoning or effort setting in the leaderboard's model name before comparing cost or score.
- Re-check the live leaderboard page before any model selection decision is approved.
- Pair leaderboard results with pricing from the LLM Pricing Tracker and your own evaluation set.
Standards and references
- Artificial Analysis: Models leaderboard (accessed 2026-11-15)
- Artificial Analysis: Intelligence benchmarking methodology (accessed 2026-11-15)
- Epoch AI: Epoch Capabilities Index (accessed 2026-11-15)
- Epoch AI: ECI model scores dataset (accessed 2026-11-15)
- Epoch AI: Benchmarking hub (accessed 2026-11-15)
- xAI: Grok 4.6 announcement (accessed 2026-11-15)
Related NHI Mgmt Group resources: How to Read LLM Benchmarks, Frontier AI Models Compared, LLM Release Tracker, LLM Pricing Tracker, Grok 4.7 vs 4.6 vs 4.5.