The full scored language board, including the specialised products listed under Other. Official publisher releases only; quantizations and repacks stay discoverable, unranked.
1Current-model rank | Anthropic | 97.12 public benchmark observations across 2 boards; Index basis Measured; 22% support; score interval 72.4–100.0. | 1M 128K max output Exact context: 1,000,000 tokens; exact maximum output: 128,000 tokens | — No Arena rating | $20.00 |
2Current-model rank | GPT-6 SolNew OpenAI | 88.41 public benchmark observation across 1 board; Index basis Measured; 20% support; score interval 67.3–100.0. | 1M 128K max output Exact context: 1,050,000 tokens; exact maximum output: 128,000 tokens | — No Arena rating | $10.00 |
3Current-model rank | Anthropic | 87.85 public benchmark observations across 5 boards; Index basis Measured; 78% support; score interval 80.6–95.0. | 1M 128K max output Exact context: 1,000,000 tokens; exact maximum output: 128,000 tokens | 1,498 LMArena Highest supported configuration: Max; source rank 5 | $50.00 |
4Current-model rank | Grok 4.7New xAI | 87.41 public benchmark observation across 1 board; Index basis Measured; 20% support; score interval 66.3–100.0. | 500K 450K max output Exact context: 500,000 tokens; exact maximum output: 450,000 tokens | — No Arena rating | $4.80 |
5Current-model rank | XiaomiLocal model | 87.31 public benchmark observation across 1 board; Index basis Measured; 20% support; score interval 66.2–100.0. | 1M 131K max output Exact context: 1,048,576 tokens; exact maximum output: 131,072 tokens | — No Arena rating | $0.87 |
6Current-model rank | 86.51 public benchmark observation across 1 board; Index basis Measured; 20% support; score interval 65.4–100.0. | 1M 131K max output Exact context: 1,000,000 tokens; exact maximum output: 131,072 tokens | — No Arena rating | $6.00 | |
7Current-model rank | OpenAI | 86.26 public benchmark observations across 6 boards; Index basis Measured; 68% support; score interval 77.6–94.8. | 1M 128K max output Exact context: 1,050,000 tokens; exact maximum output: 128,000 tokens | 1,480 LMArena Highest supported configuration: Max; source rank 24 | $50.00 |
8Current-model rank | Meta | 85.64 public benchmark observations across 4 boards; Index basis Measured; 75% support; score interval 78.0–93.2. | 1M 944K max output Exact context: 1,048,576 tokens; exact maximum output: 943,718 tokens | 1,493 LMArena Highest supported configuration: Max; source rank 8 | $4.25 |
9Current-model rank | Google | 84.14 public benchmark observations across 4 boards; Index basis Measured; 75% support; score interval 76.5–91.7. | 1M 66K max output Exact context: 1,048,576 tokens; exact maximum output: 65,536 tokens | 1,493 LMArena Highest supported configuration: High; source rank 9 | $3.75 |
10Current-model rank | XiaomiLocal model | 83.9publisher-corpus-prior over 223 held-out anchors (8%); launch card against 5 resolved peers on 7 rows (92%); exceeds every named peer on 1 of 7 rows; no cross-lab optimism probe available; shrunk 0.6 toward the measured corpus; Index basis Estimated; 33% support; score interval 75.4–92.3. | 1M 131K max output Exact context: 1,048,576 tokens; exact maximum output: 131,072 tokens | — No Arena rating | $0.28 |
11Current-model rank | 83.84 public benchmark observations across 4 boards; Index basis Measured; 83% support; score interval 77.1–90.5. | 1.3M 131K max output Exact context: 1,310,720 tokens; exact maximum output: 131,072 tokens | 1,483 LMArena Highest supported configuration: Max; source rank 19 | $2.64Increased by $0.59 | |
12Current-model rank | Moonshot AILocal model | 83.56 public benchmark observations across 6 boards; Index basis Measured; 89% support; score interval 77.4–89.6. | 1M 944K max output Exact context: 1,048,576 tokens; exact maximum output: 943,718 tokens | 1,485 LMArena Highest supported configuration: Max; source rank 17 | $15.00 |
13Current-model rank | OpenAI | 82.27 public benchmark observations across 7 boards; Index basis Measured; 90% support; score interval 76.2–88.2. | 1M 128K max output Exact context: 1,050,000 tokens; exact maximum output: 128,000 tokens | 1,482 LMArena Highest supported configuration: High; source rank 20 | $30.00 |
14Current-model rank | 81.64 public benchmark observations across 4 boards; Index basis Measured; 82% support; score interval 74.9–88.3. | 1.3M 944K max output Exact context: 1,310,720 tokens; exact maximum output: 943,718 tokens | 1,475 LMArena Highest supported configuration: Default; source rank 29 | $0.50 | |
15Current-model rank | 81.51 public benchmark observation across 1 board; Index basis Measured; 20% support; score interval 60.4–100.0. | 1M 131K max output Exact context: 1,000,000 tokens; exact maximum output: 131,072 tokens | — No Arena rating | $0.47 | |
16Current-model rank | MeituanLocal model | 81.4global-corpus-prior over 223 held-out anchors (6%); launch card against 4 resolved peers on 6 rows (94%); no cross-lab optimism probe available; shrunk 1.6 toward the measured corpus; Index basis Estimated; 32% support; score interval 73.9–88.8. | 1M 262K max output Exact context: 1,048,756 tokens; exact maximum output: 262,144 tokens | — No Arena rating | $1.20 |
17Current-model rank | Google | 81.24 public benchmark observations across 4 boards; Index basis Measured; 86% support; score interval 74.9–87.5. | 1M 66K max output Exact context: 1,048,576 tokens; exact maximum output: 65,536 tokens | 1,487 LMArena Highest supported configuration: Default; source rank 15 | $12.00 |
18Current-model rank | 81.02 public benchmark observations across 2 boards; Index basis Measured; 62% support; score interval 76.0–86.0. | 2M 1.8M max output Exact context: 2,000,000 tokens; exact maximum output: 1,800,000 tokens | 1,470 LMArena Highest supported configuration: Beta · 0309; source rank 40 | $2.50 | |
19Current-model rank | DeepSeekLocal model | 80.22 public benchmark observations across 2 boards; Index basis Measured; 23% support; score interval 59.3–100.0. | 1M 944K max output Exact context: 1,048,576 tokens; exact maximum output: 943,718 tokens | — No Arena rating | $0.50Decreased by $0.10 |
20Current-model rank | Nex AGILocal model | 80.1global-corpus-prior over 223 held-out anchors (6%); launch card against 6 resolved peers on 5 rows (94%); optimism haircut 0 from cross-lab-probe; shrunk 1.6 toward the measured corpus; Index basis Estimated; 27% support; score interval 72.7–87.6. | — — max output Exact context: 0 tokens | — No Arena rating | Self-hosted: weights only · no listed API price |
21 Increased by 2 Current-model rank | ByteDance Seed | 79.6global-corpus-prior over 223 held-out anchors (11%); launch card against 2 resolved peers on 3 rows (90%); exceeds every named peer on 1 of 3 rows; no cross-lab optimism probe available; shrunk 2.6 toward the measured corpus; Index basis Estimated; 23% support; score interval 69.9–89.2. | 262K 236K max output Exact context: 262,144 tokens; exact maximum output: 235,929 tokens | — No Arena rating | $2.50 |
22 Decreased by 1 Current-model rank | OpenAI | 79.56 public benchmark observations across 6 boards; Index basis Measured; 89% support; score interval 73.4–85.6. | 1M 128K max output Exact context: 1,050,000 tokens; exact maximum output: 128,000 tokens | 1,466 LMArena Highest supported configuration: X-High; source rank 45 | $12.00 |
23 Decreased by 1 Current-model rank | Nex AGILocal model | 79.5global-corpus-prior over 223 held-out anchors (8%); launch card against 8 resolved peers on 8 rows (92%); exceeds every named peer on 1 of 8 rows; optimism haircut 0 from cross-lab-probe; shrunk 1.9 toward the measured corpus; Index basis Estimated; 38% support; score interval 71.2–87.8. | 262K 236K max output Exact context: 262,144 tokens; exact maximum output: 235,929 tokens | — No Arena rating | $0.25 |
24Current-model rank | Ornith AiLocal model | 79.2global-corpus-prior over 223 held-out anchors (10%); launch card against 2 resolved peers on 11 rows (90%); exceeds every named peer on 3 of 11 rows; optimism haircut 0 from cross-lab-probe; shrunk 2.4 toward the measured corpus; Index basis Estimated; 40% support; score interval 69.9–88.5. | — — max output Exact context: 0 tokens | — No Arena rating | Self-hosted: weights only · no listed API price |
25Current-model rank | DeepSeekLocal model | 78.53 public benchmark observations across 3 boards; Index basis Measured; 60% support; score interval 73.3–83.7. | 1M 384K max output Exact context: 1,048,576 tokens; exact maximum output: 384,000 tokens | 1,463 LMArena Highest supported configuration: High; source rank 50 | $1.39Decreased by $0.59 |
The first ten positions on the ModelCap Index as of 23 September 2026, with the listed output price and context window beside each. The full board above is searchable and re-orders itself on every refresh.
Public evidence only. Every position comes from named public benchmark boards — Arena text and coding, LMArena Agent, ARC-AGI, SWE-bench, BFCL and others — matched to an exact model identity. Downloads, hype and provider counts never move a rank.
Uncertainty is published. Each score carries an interval and an evidence label (measured, inherited, estimated), so a one-point gap between neighbours reads as what it is. The methodology and the public dataset let anyone re-derive the board.
Rank, price and availability together. Listed API prices, context windows, weight access and every provider’s endpoint are read from the same snapshot as the rank and re-render within a minute of each refresh; the changes feed records every movement.
Claude Opus 5.5 from Anthropic holds ModelCap Index position #1 as of 23 September 2026 with a score of 97.1, ahead of GPT-6 Sol (#2) and Claude Fable 5.1 (#3). The Index combines public benchmark boards with published uncertainty, so a narrow gap between neighbours is shown as one; the ordering re-renders within a minute of every refresh.
Every current, canonical language model with matched public benchmark evidence receives a ModelCap Index score (0–100) with an interval. Evidence comes from named public boards — Arena text and coding, LMArena Agent, ARC-AGI, SWE-bench, BFCL and others — and each result counts by its board's share of its family and by its confidence, which falls as the evidence ages. A model not yet measured on those boards can hold a modeled position, labelled as such and shown with its confidence and interval. The methodology page publishes the weighting, the placement and identity rules and the change log. Market signals (downloads, provider counts) never move a rank.
Kimi K3 from Moonshot AI leads ModelCap's best-for-coding list as of 23 September 2026 (ModelCap Index #12). The list ranks models measured on at least two admitted coding boards (Arena Coding, SWE-bench bash-only, Terminal-Bench 2.1 with Terminus 2) by the Index's coding component, then lists models measured on a single board by that board's score, with the figures beside each row.
Among ranked models with a listed price, Ling 3.0 Flash is the cheapest by blended token price as of 23 September 2026 ($0.063 per 1M output tokens), at ModelCap Index #78. The cheapest-LLMs list orders every priced model that way, and 14 ranked models currently have a free tier on at least one provider.
MiMo-V2.6-Pro from Xiaomi is the highest-ranked model with openly downloadable weights as of 23 September 2026, at ModelCap Index #5. The open-weight list orders every such model by Index position with its licence class and listed price; ModelCap labels weight availability, not whether the full training stack is open source.
Grok 4.20 Multi-Agent lists the largest published context window among ranked models as of 23 September 2026: 2M tokens. The context-window list orders every ranked model by published limit; individual providers can serve less than the maximum.
The dataset refreshes continuously and every page re-renders within a minute of a new snapshot; this one was published 23 September 2026. New models appear as soon as they pass automatic identity checks, and every rank or price movement is recorded on the changes feed.
2262 canonical models across language, image, audio and video catalogues, 274 of them with a public ModelCap Index position, served by 79 inference providers observed through OpenRouter as of 23 September 2026. The models directory lists every one, A–Z by publisher.
