LIVE
Anthropic files IPO prospectus, with risk disclosures taking center stage26/10/26 · Anthropic|OpenAI models are coming to Google Cloud21/10/26 · OpenAI|OpenAI details ChatGPT's new advertising infrastructure21/10/26 · OpenAI|Anthropic reportedly posted a net loss of over $80 billion in H1 202620/10/26 · Anthropic|Self-Evolving Interfaces Help a Mid-Sized Open Model Beat GPT-5.5 and Claude Opus20/10/26|OpenAI publishes a page announcing 'GPT-6.1 Terra', with no technical details available26/10/26 · OpenAI|Cohere flags concerning behavior in a popular open-weight AI model26/10/26 · Cohere|Brown student maps the hidden global labor powering AI26/10/26|OpenAI releases data on ChatGPT's professional use across 100 countries26/10/26 · OpenAI|Nikon photography contest winner stripped of title after AI detection25/10/26|Cerebras Details Its Infrastructure for Trillion-Parameter Model Deployment25/10/26 · Cerebras|US Creative Jobs Decline by 200,000 — How Much Can Be Blamed on AI?25/10/26|Anthropic files IPO prospectus, with risk disclosures taking center stage26/10/26 · Anthropic|OpenAI models are coming to Google Cloud21/10/26 · OpenAI|OpenAI details ChatGPT's new advertising infrastructure21/10/26 · OpenAI|Anthropic reportedly posted a net loss of over $80 billion in H1 202620/10/26 · Anthropic|Self-Evolving Interfaces Help a Mid-Sized Open Model Beat GPT-5.5 and Claude Opus20/10/26|OpenAI publishes a page announcing 'GPT-6.1 Terra', with no technical details available26/10/26 · OpenAI|Cohere flags concerning behavior in a popular open-weight AI model26/10/26 · Cohere|Brown student maps the hidden global labor powering AI26/10/26|OpenAI releases data on ChatGPT's professional use across 100 countries26/10/26 · OpenAI|Nikon photography contest winner stripped of title after AI detection25/10/26|Cerebras Details Its Infrastructure for Trillion-Parameter Model Deployment25/10/26 · Cerebras|US Creative Jobs Decline by 200,000 — How Much Can Be Blamed on AI?25/10/26|

nAIvigate Barometer

The AI model market in October 2026

209 models tracked, 145 of them language models. Figures computed on October 15 from verified model sheets and the daily history, with no LLM involved. Free to reuse with attribution (CC BY 4.0).

Released this month
3
among tracked models
Median output price
6.00 $
per million tokens
Self-hostable
70 %
101 of 145 LLMs
Top of the ranking
1525Elo
Gemini 4 Argon

The month in brief

  • 3 tracked models were released in October 2026.
  • An LLM via API costs a median 2.00 $ input and 6.00 $ output per million tokens. Half of the 43 models with a verified public price charge between 3.75 $ and 15.00 $ for output.
  • 101 of 145 LLMs can run on your own infrastructure. In the cloud, 17 guarantee EU data residency and 12 offer it on request.
  • For the 44 proprietary LLMs, we were able to verify 83% of our criteria with the vendor on average. Enterprise requirements (Vendor private endpoint, Subprocessor list published, SCIM provisioning) are the least verified so far: collection is under way.

This month's ranking

LMArena text Elo, among tracked models.

First when computed on October 15: Gemini 4 Argon (Elo 1525). No new entry in the top 10.

  1. 1Gemini 4 Argon1525 =
  2. 2Claude Opus 5.51507 ▲ from 4
  3. 3Claude Fable 51504 ▼ from 2
  4. 4Claude Opus 4.61504 ▼ from 3
  5. 5Claude Fable 5.11501 ▲ from 6
  6. 6Claude Opus 4.71501 ▼ from 5
  7. 7Gemini 3.8 Flash1497 ▲ from 8
  8. 8Muse Spark 1.31494 ▼ from 7
  9. 9Muse Spark 1.11491 =
  10. 10Claude Opus 51490 =

What an LLM costs via API

Computed on the 43 LLMs whose public price was checked with the vendor, out of 145 tracked. Models without a public price or self-hosted only are left out.

Median input price
2.00 $per million tokens
Half of the models between 0.77 $ and 3.50 $
Median output price
6.00 $per million tokens
Half of the models between 3.75 $ and 15.00 $

Cheapest (input + output)

  1. 1Claude Haiku 5.50.10 $ / 0.50 $
  2. 2GPT-6 Luna0.10 $ / 0.50 $
  3. 3Mistral Small 40.15 $ / 0.60 $
  4. 4Codestral0.30 $ / 0.90 $
  5. 5GPT-5.6 Luna0.20 $ / 1.20 $

This month's changes

No price change recorded on the 43 models with a verified public price (14 days of snapshots).

Can your data stay in Europe

Out of the 145 language models tracked.

On your own infrastructure
101/ 145
Published weights, self-hostable
Cloud, EU residency
17/ 145
Guarantee published by the vendor
Cloud, EU on request
12/ 145
Usually an enterprise contract

All modalities included (image, video, audio and 3D): 137 self-hostable models out of 209.

What we verified with proprietary vendors

For the 44 proprietary LLMs tracked: share of our 20 criteria (data, sovereignty, transparency, enterprise requirements, licence) for which we recorded information published by the vendor. A criterion not yet recorded is not necessarily missing at the vendor: collection of enterprise requirements is under way.

Share verified, on average
83 %
Criteria applicable to the model
More than half verified
36/ 44
The others have a provisional Trust Index

Least verified criteria so far

  • Vendor private endpoint22 of 44
  • Subprocessor list published30 of 44
  • SCIM provisioning30 of 44
  • Zero Data Retention32 of 44

Each model's Trust Index is on its sheet. We do not publish an average pitting open against proprietary models: a model you host yourself earns by construction the criteria you hold, which would make the comparison misleading. See the Trust Index

Released in October 2026 (3)

Official vendor release date, among tracked models.

Added to our catalogue this month (29)

Models released earlier, added to our tracking during the month (date added).

Three podiums, three winners

What users prefer, what models can do and what the market uses do not point to the same models.

Preference (LMArena Elo)
  1. 1Gemini 4 Argon1525
  2. 2Claude Opus 5.51507
  3. 3Claude Fable 51504
  4. 4Claude Opus 4.61504
  5. 5Claude Fable 5.11501
Capability (Artificial Analysis index)
  1. 1Claude Opus 5.557.6
  2. 2Claude Sonnet 5.556.0
  3. 3Muse Wonder Full55.8
  4. 4Claude Fable 5.153.4
  5. 5GPT-6 Astra52.7
Adoption (share of OpenRouter tokens)
  1. 1DeepSeek V4.1-Flash20.7 %
  2. 2space-bunny-alpha8.1 %
  3. 3GLM-5.3 Flash6.8 %
  4. 4MiMo-V2.6 Flash6.3 %
  5. 5Step 5 Preview5.5 %

LMArena (lmarena.ai) · Source: Artificial Analysis, as of 2026-10-15 · Source: OpenRouter (openrouter.ai/rankings), as of 2026-10-15. CC BY 4.0

What the market uses

Share of tokens processed by OpenRouter from 2026-10-01 to 2026-10-14, and change versus the previous month.

DeepSeek V4.1-Flash
20.7 % (+11.7)
space-bunny-alpha
8.1 % (+3.2)
GLM-5.3 Flash
6.8 % (-3.1)
MiMo-V2.6 Flash
6.3 % (+4.6)
Step 5 Preview
5.5 % (+5.5)
DeepSeek V4-Flash
4.8 % (-5.7)
Hy4 preview
4.2 % (-5.5)
GPT-6 Luna
3.9 % (+3)
Nemotron 3 Ultra 550B-A55B
3.9 % (+0.6)
Claude Opus 5.5
2.6 % (+2.3)

In brackets: change in points versus the previous month.

By vendor
DeepSeek
26.6 %
Z.ai
9.8 %
OpenAI
8.3 %
Anonymous models
8.1 %
Xiaomi
7.5 %
Tencent
5.6 %
StepFun
5.5 %
Anthropic
5.4 %
Others
23.2 %

Source: OpenRouter (openrouter.ai/rankings), as of 2026-10-15. CC BY 4.0

Sovereignty of usage

Origin of the models processing the observed tokens: actual dependency, not just available supply.

United States 23.5 %China 58.2 %Europe 0.03 %Other countries 0.9 %Unidentified (anonymous models) 11 %Long tail not broken down by OpenRouter 6.4 %

Source: OpenRouter (openrouter.ai/rankings), as of 2026-10-15. CC BY 4.0

Most used agents and apps

Apps ranked by tokens consumed through OpenRouter: the most direct signal of what teams use day to day.

1Hermes Agent
25.4 T
2Claude Code
18.5 T
3Kilo Code
12.7 T
4Cline
10 T
5Codex
7.3 T
6Freebuff
7 T
7pi
6 T
8omp
5.1 T
9OpenClaw
4.4 T
10DeepSeek Harness
2.9 T

Source: OpenRouter (openrouter.ai/rankings), as of 2026-10-15. CC BY 4.0

Which model for which job

Usage split by task type and the most used models for each (7-day window).

Classification
31.3 % of requests
  1. jev-1.1330.5 %
  2. DeepSeek V4-Flash6 %
  3. DeepSeek V4.1-Flash4.9 %
Data Extraction
12.5 % of requests
  1. DeepSeek V4-Flash9.6 %
  2. DeepSeek V4.1-Flash7.5 %
  3. GLM-5.3 Flash7.1 %
Workflow Execution
8.8 % of requests
  1. DeepSeek V4.1-Flash18.7 %
  2. GLM-5.3 Flash10.2 %
  3. DeepSeek V4-Flash5.7 %
Roleplay & Fiction
7.2 % of requests
  1. DeepSeek V4-Flash25.8 %
  2. DeepSeek V4.1-Flash13.5 %
  3. mistral-nemo4.5 %
Content Writing
5.1 % of requests
  1. GLM-5.3 Flash25.8 %
  2. DeepSeek V4-Flash14.9 %
  3. DeepSeek V4.1-Flash10.8 %
Data Transformation
4.6 % of requests
  1. GLM-5.3 Flash30 %
  2. DeepSeek V4.1-Flash12 %
  3. DeepSeek V4-Flash10.3 %
Code Generation
3.7 % of requests
  1. DeepSeek V4.1-Flash32.2 %
  2. MiMo-V2.6 Flash8 %
  3. GLM-5.3 Flash7.4 %
Q&A & Knowledge
3.5 % of requests
  1. jev-1.1315.1 %
  2. GPT-6 Luna9.9 %
  3. DeepSeek V4.1-Flash9.2 %
Multi-step Planning
3.1 % of requests
  1. DeepSeek V4.1-Flash18.8 %
  2. GLM-5.3 Flash11.7 %
  3. DeepSeek V4-Flash6 %
Conversation
2.3 % of requests
  1. DeepSeek V4-Flash16 %
  2. DeepSeek V4.1-Flash10.2 %
  3. GLM-5.3 Flash5 %

Source: OpenRouter, as of 2026-10-14. CC BY 4.0

Capability versus price

Each dot is a model tracked by nAIvigate. The line links the best trade-offs: no model is both cheaper and more capable.

01020304050600.10.3131030Average price ($ per million tokens, 3 input to 1 output, log scale)Artificial Analysis indexClaude Opus 5.5 · 57.6 · 8 $Claude Sonnet 5.5 · 56 · 4 $Claude Fable 5.1 · 53.4 · 20 $GPT-6 Astra · 52.7 · 20 $Gemini 4 Argon · 52.6 · 4 $GPT-6.1 Sol · 51.8 · 4 $Claude Opus 5 · 50.8 · 10 $Claude Fable 5 · 49.6 · 20 $Muse Spark 1.3 · 48.1 · 2 $GPT-6 Sol · 47.6 · 4 $GPT-5.6 Sol · 47 · 8 $Grok 4.7 · 46.4 · 3 $MiMo-V2.6 Pro · 46.3 · 0.544 $Qwen3.8-Max · 45.4 · 3 $GLM-5.3 · 44.8 · 2.15 $Step 5 Preview · 43.7 · 1.425 $Kimi K3 · 43.6 · 6 $Claude Haiku 5.5 · 43.4 · 0.2 $GPT-5.6 Terra · 42.1 · 4.5 $GLM-5.3 Flash · 41.8 · 0.237 $Claude Opus 4.8 · 41.8 · 10 $Gemini 3.8 Flash · 40.9 · 1.5 $Claude Opus 4.7 · 40.7 · 10 $Gemini 3.7 Flash · 39.6 · 1.5 $DeepSeek V4.1-Flash · 39.5 · 0.525 $Grok 4.5 · 38.8 · 3 $Mistral Large 4 · 38.4 · 2.065 $GPT-5.5 · 38.4 · 11.25 $Claude Sonnet 5 · 38.2 · 4 $GPT-6 Luna · 38.1 · 0.2 $MiMo-V2.6 Flash · 37.9 · 0.175 $GPT-5.6 Luna · 37.3 · 0.45 $DeepSeek V4 Pro · 36 · 1.98 $DeepSeek V4-Flash · 34.3 · 0.66 $Gemini 3.6 Flash · 34 · 1.5 $Muse Spark 1.1 · 33.7 · 2 $GLM-5.2 · 33.7 · 2.15 $Qwen3.8 27B · 33.7 · 1.125 $Claude Opus 4.6 · 31.9 · 10 $Claude Sonnet 4.6 · 30.1 · 6 $Gemini 3.1 Pro · 29.7 · 4.5 $Qwen3.7 Max · 29.5 · 3.75 $MiniMax M3 · 29.2 · 0.525 $Gemini 3 Pro · 28 · 4.5 $GLM-5 · 27.9 · 1.55 $Kimi K2.6 · 27 · 1.712 $Quasar 438B · 26.7 · 0.9 $Gemini 3 Flash · 26.3 · 1.125 $Kimi K2.7 Code · 25.8 · 1.712 $Hy3 · 25.3 · 0.241 $Qwen3.7 Plus · 25.2 · 0.7 $GPT-5 · 23 · 3.438 $Nemotron 3 Ultra 550B-A55B · 22.9 · 1.075 $Grok 4 · 22.5 · 6 $Kimi K2 · 22 · 1.075 $DeepSeek V3.2 · 21.5 · 0.315 $Qwen3.5 397B-A17B · 21.4 · 1.35 $Qwen3.6 27B · 21.4 · 1.35 $GPT-5 mini · 20.6 · 0.688 $o3 · 20.2 · 3.5 $Qwen3.6 35B-A3B · 18.2 · 0.844 $Gemma 4 26B-A4B · 16.7 · 0.175 $Mistral Medium 3.5 · 14.2 · 3 $Gemma 4 12B · 14.2 · 0.15 $DeepSeek-R1 · 13.1 · 1.763 $Nemotron 3 Super 120B-A12B · 12.8 · 0.45 $gpt-oss-120b · 11.6 · 0.26 $Mistral Small 4 · 11.3 · 0.262 $GLM-4.5 Air · 11.1 · 0.372 $gpt-oss-20b · 10 · 0.106 $Llama 4 Maverick · 9.3 · 0.405 $Mistral Large 3 · 9.3 · 0.75 $Gemma 4 E4B · 8.9 · 0.04 $DeepSeek V3 · 8.5 · 0.463 $Qwen2.5 72B · 7.7 · 0.48 $Llama 3.3 70B · 7.7 · 0.712 $Qwen3 32B · 7.3 · 0.262 $Mistral Small 3.2 · 7 · 0.106 $Llama 3.1 8B · 6.9 · 0.043 $Qwen3 14B · 6.7 · 0.15 $Mistral Small 3 · 6.7 · 0.058 $Llama 4 Scout · 6.5 · 0.3 $Phi-4 Mini · 6.3 · 0.131 $Ministral 3 14B · 6 · 0.2 $Qwen3 8B · 6 · 0.2 $Phi-4 · 5.9 · 0.219 $Ministral 3 8B · 5.5 · 0.15 $OLMo 3 7B · 5.2 · 0.125 $Mixtral 8x7B · 5.1 · 0.512 $Mistral 7B · 5 · 0.162 $Granite 3.3 8B · 4.9 · 0.085 $Ministral 3 3B · 4.8 · 0.1 $Apertus 8B · 4.8 · 0.125 $Gemma 4 E4Bgpt-oss-20bGemma 4 12BMiMo-V2.6 FlashClaude Haiku 5.5MiMo-V2.6 ProMuse Spark 1.3Claude Sonnet 5.5Claude Opus 5.5
Open-weightProprietary

Source: Artificial Analysis, as of 2026-10-15

The gap between open and closed models

Best index available at each month end, among models tracked by nAIvigate. Current index applied retroactively: older models were not rated on this version.

1020304050602025-012025-042025-072025-102026-012026-042026-072026-10

Today : Claude Opus 5.5 (57.6) · MiMo-V2.6 Pro (46.3) · 11.3 points apart

ProprietaryOpen-weight

Source: Artificial Analysis, as of 2026-10-15

The fastest

Output tokens per second, median measured on the vendor API. Speed also depends on the host.

Celeris-1
1779 tokens/s
Mercury 2.5
724 tokens/s
K2 Horizon 375B A23B
407 tokens/s
Gemini 3.5 Flash-Lite
406 tokens/s
Trinity Large Thinking
362 tokens/s
Nemotron 3.5 Lightning
300 tokens/s
gpt-oss-20b
292 tokens/s
Ling-3.0-flash-Fin
272 tokens/s
Nova Micro
250 tokens/s
Nemotron 3 Nano Omni 30B A3B Reasoning
241 tokens/s

Source: Artificial Analysis, as of 2026-10-15

Method

Language models: all tracked models except image, video, music, voice and 3D generation. Prices in US dollars per million tokens, collected from vendor pricing pages; median and middle range (25th to 75th percentile) over LLMs whose input and output prices are public and verified; corrections to our own data are flagged separately and do not count as price changes. Releases: official release date within the month. Additions: older models added to the catalogue during the month. Arena ranking: LMArena text Elo as recorded when the figures are computed, compared with the ranking at the month's first daily snapshot. Models released during the month count as entries; models we rate for the first time because we just added their Arena mapping do not. A change below 10 Elo points between the first and last snapshot is not counted as a move. EU residency: guarantee published by the vendor for cloud hosting inside the European Union. Verification: applicable Trust Index criteria for which we recorded information published by the vendor; a criterion not recorded counts as missing, whether unpublished or not yet verified. Variations computed over 269 snapshots, from October 2 to October 15. No placement is sold, no figure is produced by an LLM.

Cite: "nAIvigate Barometer, October 2026", https://naivigate.com/en/barometre/2026-10

Archives: 2026-102026-09