Week 45, updated 2 Nov 2026

AI models

18 frontier and major models from official pages and 15 open models from Hugging Face. Higher and further left is more capable for less.

  • Closed weights
  • Open or announced open weights
  • Highest index: Gemini 4 Carbon

Each mark is one model: capability (Artificial Analysis Intelligence Index, up) against output price (right).

Ranked by capability (AA Intelligence Index), with the output price.

17 models plotted. Index from the Artificial Analysis LLM Leaderboard (Intelligence Index v5.0, launched 27 Oct 2026), retrieved 2 Nov 2026; prices from each provider's pricing page, verified 2 Nov 2026. Not plotted because a value is missing: Muse Spark Open (no API price).

$0.5$1$2$5$10$20$5020253035404550556065 Output price, USD per 1M tokens (log scale)AA Intelligence Index OpenAI GPT-6 Astra: index 57 (max), $10 in / $50 out per 1M tokensGPT-6 AstraOpenAI GPT-6.1 Sol: index 55 (max), $2 in / $10 out per 1M tokensGPT-6.1 SolOpenAI GPT-6 Luna: index 28 (max), $0.1 in / $0.5 out per 1M tokensGPT-6 LunaOpenAI GPT-6.1 Terra: index 51 (max), $1 in / $5 out per 1M tokensGPT-6.1 TerraAnthropic Claude Opus 5.5: index 58 (max), $4 in / $20 out per 1M tokensClaude Opus 5.5Anthropic Claude Sonnet 5.5: index 54 (max), $2 in / $10 out per 1M tokensClaude Sonnet 5.5Anthropic Claude Haiku 5.5: index 36 (max), $0.1 in / $0.5 out per 1M tokensClaude Haiku 5.5Anthropic Claude Fable 5.1: index 50 (xhigh), $10 in / $50 out per 1M tokensClaude Fable 5.1Google Gemini 4 Carbon: index 59 (high), $2 in / $10 out per 1M tokensGemini 4 CarbonGoogle Gemini 4 Argon: index 50 (high), $2 in / $10 out per 1M tokensGemini 4 ArgonGoogle Gemini 3.8 Flash: index 31 (high), $0.75 in / $3.75 out per 1M tokensGemini 3.8 FlashxAI Grok 4.7: index 35 (high), $2 in / $6 out per 1M tokensGrok 4.7Meta Muse Wonder: index 43 (max), $4 in / $20 out per 1M tokensMuse WonderMistral Mistral Large 4 (preview): index 26 (high), $1.36 in / $4.18 out per 1M tokensMistral Large 4 (prev.)DeepSeek DeepSeek V4.1 Flash 1017: index 32 (max), $0.3 in / $1.2 out per 1M tokensDeepSeek V4.1 FlashQwen (Alibaba) Qwen3.8-Max / Qwen3.8-2.4T-A95B: index 34 (Qwen3.8 Max 0902), $2 in / $6 out per 1M tokensQwen3.8-MaxMoonshot AI (Kimi) Kimi K3.1 Pro: index 46, $2 in / $10 out per 1M tokensKimi K3.1 Pro
AA Intelligence Indexoutput price per 1M tokens
  1. Gemini 4 CarbonGoogle, $10 out59
  2. Claude Opus 5.5Anthropic, $20 out58
  3. GPT-6 AstraOpenAI, $50 out57
  4. GPT-6.1 SolOpenAI, $10 out55
  5. Claude Sonnet 5.5Anthropic, $10 out54
  6. GPT-6.1 TerraOpenAI, $5 out51
  7. Gemini 4 ArgonGoogle, $10 out50
  8. Claude Fable 5.1Anthropic, $50 out50
  9. Kimi K3.1 ProMoonshot AI (Kimi), $10 out, open weights announced46
  10. Muse WonderMeta, $20 out43
  11. Claude Haiku 5.5Anthropic, $0.5 out36
  12. Grok 4.7xAI, $6 out35
  13. Qwen3.8-Max / Qwen3.8-2.4T-A95BQwen (Alibaba), $6 out, open weights34
  14. DeepSeek V4.1 Flash 1017DeepSeek, $1.2 out, open weights32
  15. Gemini 3.8 FlashGoogle, $3.75 out31
  16. GPT-6 LunaOpenAI, $0.5 out28
  17. Mistral Large 4 (preview)Mistral, $4.18 out, open weights announced26
Read this before comparing

Providers are listed at different reasoning settings (max, high, xhigh, preview), so small gaps are not meaningful. Prices are API list prices in USD per 1M output tokens (tier notes are in the table). Every value links to its source in the full table below. We do not run benchmarks ourselves.

Released in the last 60 days

14 frontier models.

  1. Google Gemini 4 Carbon $2 / $10Closed
  2. OpenAI GPT-6.1 Terra 1.1M context, $1 / $5Closed
  3. Moonshot AI (Kimi) Kimi K3.1 Pro 1M context, $2 / $10Open announced
  4. DeepSeek DeepSeek V4.1 Flash 1017 1M context, $0.3 / $1.2Open weights
  5. Meta Muse Spark Open 1M contextOpen weights
  6. Meta Muse Wonder 1M context, $4 / $20Closed
  7. Anthropic Claude Haiku 5.5 1M context, $0.1 / $0.5Closed
  8. Mistral Mistral Large 4 (preview) 1M context, $1.36 / $4.18Open announced
  9. Google Gemini 4 Argon 1M context, $2 / $10Closed
  10. OpenAI GPT-6.1 Sol 1.1M context, $2 / $10Closed
  11. Anthropic Claude Sonnet 5.5 1M context, $2 / $10Closed
  12. OpenAI GPT-6 Luna 1.1M context, $0.1 / $0.5Closed
  13. Anthropic Claude Opus 5.5 1M context, $4 / $20Closed
  14. xAI Grok 4.7 500k context, $2 / $6Closed

Most downloaded open models

Hugging Face downloads, last 30 days.

Downloads, 30 daysHugging Face API
  1. 1Qwen3.8-27BQwen, apache-2.07.9M
  2. 2Qwen3.8-Flash-NextQwen, other4M
  3. 3gpt-oss-120bopenai, apache-2.02.5M
  4. 4LTX-2.5Lightricks, other2.4M
  5. 5DeepSeek-V4.1-Flashdeepseek-ai, mit2.1M
  6. 6MiniCPM-V-4.7-35B-A3Bopenbmb, apache-2.0423k
  7. 7embeddinggemma-2google, apache-2.0388k
  8. 8gemma-3-27b-itgoogle, gemma351k
All 15 open models, with the facts
Open models, official Hugging Face data
ModelWeights / licenseReleasedContext (tokens)API price / 1M tokensinput / outputModalitiesBenchmarksSources
Qwen/Qwen3.8-27BQwen · Hugging FaceOpen weightsapache-2.0262,144self-host–7,892,866 downloads (30d)[1]
Qwen/Qwen3.8-Flash-NextQwen · Hugging FaceOpen weightsother262,144self-host–4,029,520 downloads (30d)[1]
openai/gpt-oss-120bopenai · Hugging FaceOpen weightsapache-2.0131,072self-host–2,549,284 downloads (30d)[1]
Lightricks/LTX-2.5Lightricks · Hugging FaceOpen weightsother–self-host–2,431,846 downloads (30d)[1]
deepseek-ai/DeepSeek-V4.1-Flashdeepseek-ai · Hugging FaceOpen weightsmit1,048,576self-host–2,127,615 downloads (30d)[1]
openbmb/MiniCPM-V-4.7-35B-A3Bopenbmb · Hugging FaceOpen weightsapache-2.0262,144self-host–422,545 downloads (30d)[1]
google/embeddinggemma-2google · Hugging FaceOpen weightsapache-2.0262,144self-host–387,707 downloads (30d)[1]
google/gemma-3-27b-itgoogle · Hugging FaceOpen weightsgemma–self-host–351,339 downloads (30d)[1]
moonshotai/Kimi-K2-Instructmoonshotai · Hugging FaceOpen weightsother131,072self-host–314,569 downloads (30d)[1]
deepseek-ai/DeepSeek-V3.1deepseek-ai · Hugging FaceOpen weightsmit163,840self-host–271,373 downloads (30d)[1]
Qwen/Qwen-Image-2.1Qwen · Hugging FaceOpen weightsother–self-host–217,429 downloads (30d)[1]
mistralai/Mistral-Small-3.2-24B-Instruct-2506mistralai · Hugging FaceOpen weightsapache-2.0131,072self-host–122,835 downloads (30d)[1]
Qwen/Qwen3-235B-A22B-Instruct-2507Qwen · Hugging FaceOpen weightsapache-2.0262,144self-host–107,507 downloads (30d)[1]
meta-llama/Llama-4-Scout-17B-16E-Instructmeta-llama · Hugging FaceOpen weightsother–self-host–105,698 downloads (30d)[1]
LiquidAI/LFM2.5-24B-A2BLiquidAI · Hugging FaceOpen weightsother131,072self-host–30,495 downloads (30d)[1]

Every value with its source

Empty cells (–) mean no official figure was found.

Full table, 18 frontier models
Show:

Frontier and major models

Major models, official data and public benchmarks
ModelWeights / licenseReleasedContext (tokens)API price / 1M tokensinput / outputModalitiesBenchmarksSources
GPT-6 AstraOpenAI · gpt-6-astraClosedProprietary1,050,000$10 / $50Standard tier, ≤272K inputText, image → textAA Intelligence Index (max)57[1] [2] [3]
GPT-6.1 SolOpenAI · gpt-6.1-solClosedProprietary1,050,000$2 / $10Standard tier, ≤272K inputText, image → textAA Intelligence Index (max)55[1] [2] [3]
GPT-6 LunaOpenAI · gpt-6-lunaClosedProprietary1,050,000$0.10 / $0.50Standard tier, ≤272K inputText, image → textAA Intelligence Index (max)28[1] [2] [3]
GPT-6.1 TerraOpenAI · gpt-6.1-terraClosedProprietary1,050,000$1 / $5Standard tier, ≤272K inputText, image → textAA Intelligence Index (max)51[1] [2]
Claude Opus 5.5Anthropic · claude-opus-5-5ClosedProprietary1,000,000$4 / $20Text, image → textAA Intelligence Index (max)58[1] [2]
Claude Sonnet 5.5Anthropic · claude-sonnet-5-5ClosedProprietary1,000,000$2 / $10Text, image → textAA Intelligence Index (max)54[1] [2]
Claude Haiku 5.5Anthropic · claude-haiku-5-5ClosedProprietary1,000,000$0.10 / $0.50"From" price per AnthropicText, image → textAA Intelligence Index (max)36[1] [2]
Claude Fable 5.1Anthropic · claude-fable-5-1ClosedProprietary–1,000,000$10 / $50Text, image → textAA Intelligence Index (xhigh)50[1]
Gemini 4 CarbonGoogle · ClosedProprietary–$2 / $10Introductory price; $4 / $20 after the introductory period. Limited access (cyber defenders, Fairwind Program), not yet generally available–AA Intelligence Index (high)59[1]
Gemini 4 ArgonGoogle · gemini-4-argonClosedProprietary1,000,000$2 / $10Introductory until 30 Nov 2026; $4 / $20 from 1 Dec 2026. Announced 30 Sep, API GA 13 Oct–AA Intelligence Index (high)50[1] [2] [3]
Gemini 3.8 FlashGoogle · gemini-3.8-flashClosedProprietary––$0.75 / $3.75Until 31 Dec 2026; $1.50 / $7.50 from 1 Jan 2027–AA Intelligence Index (high)31[1] [2]
Grok 4.7xAI · grok-4.7ClosedProprietary500,000$2 / $6<200K prompt tokens; $4 / $12 aboveText, image → textAA Intelligence Index (high)35[1] [2] [3]
Muse WonderMeta · muse-wonder-1.0ClosedProprietary1,000,000$4 / $20Meta Model API–AA Intelligence Index (max)43[1] [2]
Muse Spark OpenMeta · meta-models/Muse-Spark-Open-816B-A42BOpen weightsMuse Model License (custom; gated download)1,048,576– / –No Meta API; self-host or via hosted providers (Baseten, Fireworks AI, Groq, Modal, Nebius, Together AI, OpenRouter)Text, image, audio, video → textAA Intelligence Index (xhigh)25[1] [2] [3]
Mistral Large 4 (preview)Mistral · mistral-large-4Open weights announcedNot yet published1,000,000$1.36 / $4.18List price; preview sale price $0.68 / $2.09Text, image → textAA Intelligence Index (high)26[1] [2]
DeepSeek V4.1 Flash 1017DeepSeek · deepseek-flashOpen weightsMIT1,000,000$0.30 / $1.20Peak, cache miss; off-peak half price. Price unchanged from V4.1 FlashText, image → textAA Intelligence Index (max)32[1] [2] [3]
Qwen3.8-Max / Qwen3.8-2.4T-A95BQwen (Alibaba) · qwen3.8-maxOpen weightsQwen3.8-Max license (custom)1,000,000$2 / $6International (Singapore) list price; Global deployment $1.65 / $4.951Text, image → textAA Intelligence Index (Qwen3.8 Max 0902)34Text Arena rank (Alibaba-reported): #5[1] [2] [3] [4]
Kimi K3.1 ProMoonshot AI (Kimi) · kimi-k3.1-proOpen weights announcedNot yet published1,048,576$2 / $10Flat price across full context; Highspeed variant (kimi-k3.1-pro-highspeed) $4 / $20. Open weights promised by 4 Nov 2026Text, image → textAA Intelligence Index46[1] [2] [3] [4]

Bars show the Artificial Analysis Intelligence Index on its 0–100 scale (higher is better); the label names the reasoning setting. Rank-type values have no bar.

Prices are API list prices in USD per 1M tokens (standard tier) from each provider's pricing page; promotional or tiered prices are noted. Benchmarks: Artificial Analysis Intelligence Index (leaderboard, retrieved 2 Nov 2026), a composite score at the listed reasoning setting, or provider-reported figures where marked. We don't run benchmarks ourselves.

Get the board every Monday

One email a week: the new top 10, the biggest score changes and every safety label that changed.

Free. Double opt-in: you confirm by email first, unsubscribe in one click.