Models

The labs keep shipping. This page is the current flagship cut: closed APIs from OpenAI, Anthropic, Google, xAI, and Meta, plus the open-weight models that actually sit next to them on a shared harness — Kimi, GLM, Qwen, DeepSeek, Gemma, Mistral, Llama. Quality is the Artificial Analysis Intelligence Index v4.1.1, not a press-release table.

Prices are USD per million tokens on the first-party (or AA-measured) API. If a field is missing, it is a dash, not a guess. Llama 4 is still here because that is the downloadable Meta line; Muse Spark is the Meta API that actually scores. We do not treat every 7B fine-tune as a frontier model.

Artificial Analysis Intelligence Index v4.1.1

Higher is better. Navy is frontier / API-only. Accent is open weight. Effort settings are the ones AA scored, not always the lab default.

Frontier closedOpen weight

Claude Opus 563
Claude Fable 562
GPT-5.6 Sol61
Grok 4.661
Kimi K360
GLM-5.360
Qwen3.8 2.4T A95B58
GPT-5.6 Terra57
Muse Spark 1.257
Gemini 3.7 Flash56
Claude Sonnet 555
DeepSeek V4 Pro 081353
DeepSeek V4 Flash 073152
GPT-5.6 Luna52
Gemini 3.1 Pro Preview48
Claude Haiku 4.530
Gemma 4 31B30
Mistral Medium 3.530
Llama 4 Maverick14

Filter

Sort

Claude Opus 5
Anthropic · Frontier closed
Anthropic's current flagship for long-horizon agents and deep reasoning.
Index
63
Context
1M
In / 1M
$5
Out / 1M
$25
License
—
Released
24 Jul 2026
Scored: adaptive reasoning, max effort
Claude Fable 5
Anthropic · Frontier closed
Highest-priced Claude in the current lineup; reserved for the heaviest agent jobs.
Index
62
Context
1M
In / 1M
$10
Out / 1M
$50
License
—
Released
9 Jun 2026
Scored: adaptive reasoning, max effort, Opus 4.8 fallback
GPT-5.6 Sol
OpenAI · Frontier closed
OpenAI's GPT-5.6 flagship; the gpt-5.6 alias routes here.
Index
61
Context
1.05M
In / 1M
$4
Out / 1M
$20
License
—
Released
9 Jul 2026
Scored: max effort
Promotional list price through at least 21 Nov 2026. Prompts over 272k input tokens bill 2× input and 1.5× output.
Grok 4.6
xAI · Frontier closed
xAI's current coding and agent flagship on a 500k window.
Index
61
Context
500k
In / 1M
$2
Out / 1M
$6
License
—
Released
12 Aug 2026
Scored: high effort
Standard band under 200k prompt tokens, per xAI docs.
Kimi K3
Moonshot / Kimi · Open weight
Highest-ranked open-weight model on the current Intelligence Index (2.8T MoE, 104B active).
Index
60
Context
1.05M
In / 1M
$3
Out / 1M
$15
License
Kimi K3 License (commercial with restrictions)
Released
16 Jul 2026
Scored: max effort
GLM-5.3
Zhipu / Z.AI · Open weight
Zhipu's current open-weight frontier; ties Kimi K3 at 60 on the Index.
Index
60
Context
1M
In / 1M
$1.40
Out / 1M
$4.40
License
GLM-5 license
Released
14 Aug 2026
Scored: max effort
Qwen3.8 2.4T A95B
Alibaba / Qwen · Open weight
Alibaba's 2.4T MoE open-weight flagship (95B active, per the AA card).
Index
58
Context
984k
In / 1M
$2
Out / 1M
$6
License
Qwen3 license
Released
12 Aug 2026
GPT-5.6 Terra
OpenAI · Frontier closed
Mid GPT-5.6 tier: most of Sol's window at half the input price.
Index
57
Context
1.05M
In / 1M
$2
Out / 1M
$12
License
—
Released
9 Jul 2026
Scored: max effort
Muse Spark 1.2
Meta · Frontier closed
Meta's API-only quality line; AA lists it proprietary, not Llama weights.
Index
57
Context
1.05M
In / 1M
$1.25
Out / 1M
$4.25
License
—
Released
5 Aug 2026
Scored: xhigh
Gemini 3.7 Flash
Google · Frontier closed
Google's current high-throughput Gemini; scores above 3.1 Pro Preview on this Index.
Index
56
Context
1M
In / 1M
$0.75
Out / 1M
$3.75
License
—
Released
13 Aug 2026
Scored: high
Claude Sonnet 5
Anthropic · Frontier closed
Workhorse Claude: 1M context at Sonnet prices.
Index
55
Context
1M
In / 1M
$2
Out / 1M
$10
License
—
Released
30 Jun 2026
Scored: adaptive reasoning, max effort
DeepSeek V4 Pro 0813
DeepSeek · Open weight
1.6T MoE, 49B active, MIT weights; the current DeepSeek quality cut.
Index
53
Context
1M
In / 1M
$1.32
Out / 1M
$3.96
License
MIT
Released
13 Aug 2026
Scored: max effort
First-party API as measured by Artificial Analysis (may differ from off-peak list).
DeepSeek V4 Flash 0731
DeepSeek · Open weight
Cheaper DeepSeek V4 cut that still clears 50 on the Index.
Index
52
Context
1M
In / 1M
$0.44
Out / 1M
$1.32
License
MIT
Released
31 Jul 2026
Scored: max effort
First-party API as measured by Artificial Analysis.
GPT-5.6 Luna
OpenAI · Frontier closed
Volume GPT-5.6 tier: same 1.05M window, priced for high-throughput work.
Index
52
Context
1.05M
In / 1M
$0.20
Out / 1M
$1.20
License
—
Released
9 Jul 2026
Scored: max effort
Gemini 3.1 Pro Preview
Google · Frontier closed
Google's Pro-class preview; Flash 3.7 now outscores it on this Index.
Index
48
Context
1M
In / 1M
—
Out / 1M
—
License
—
Released
19 Feb 2026
AA lists cost per task, not a clean input/output pair we independently confirmed on Google's pricing page.
Claude Haiku 4.5
Anthropic · Frontier closed
Fast, cheap Claude still on the current Anthropic price card (knowledge cutoff Feb 2025 on that page).
Index
30
Context
200k
In / 1M
$1
Out / 1M
$5
License
—
Released
15 Oct 2025
Gemma 4 31B
Google · Open weight
Google's current Gemma dense open-weight; Apache 2.0, 256k context.
Index
30
Context
256k
In / 1M
—
Out / 1M
—
License
Apache 2.0
Released
2 Apr 2026
AA lists $0 / $0; treat as self-host, not a Google list price.
Mistral Medium 3.5
Mistral · Open weight
Mistral's current mid-size open-weight; Large 3 scores well below this on the Index.
Index
30
Context
256k
In / 1M
$1.50
Out / 1M
$7.50
License
Modified MIT
Released
29 Apr 2026
Llama 4 Maverick
Meta · Open weight
Still the Llama-branded open-weight you can download; far behind Muse Spark on this Index.
Index
14
Context
1M
In / 1M
$0.26
Out / 1M
$0.91
License
Llama 4 Community License
Released
5 Apr 2025
Median provider price on AA, not a first-party Meta API.

	Lab	Kind					License
Claude Opus 5 Anthropic's current flagship for long-horizon agents and deep reasoning. Scored: adaptive reasoning, max effort	Anthropic	Frontier closed	63	1M	$5	$25	—	24 Jul 2026
Claude Fable 5 Highest-priced Claude in the current lineup; reserved for the heaviest agent jobs. Scored: adaptive reasoning, max effort, Opus 4.8 fallback	Anthropic	Frontier closed	62	1M	$10	$50	—	9 Jun 2026
GPT-5.6 Sol OpenAI's GPT-5.6 flagship; the gpt-5.6 alias routes here. Scored: max effort Promotional list price through at least 21 Nov 2026. Prompts over 272k input tokens bill 2× input and 1.5× output.	OpenAI	Frontier closed	61	1.05M	$4	$20	—	9 Jul 2026
Grok 4.6 xAI's current coding and agent flagship on a 500k window. Scored: high effort Standard band under 200k prompt tokens, per xAI docs.	xAI	Frontier closed	61	500k	$2	$6	—	12 Aug 2026
Kimi K3 Highest-ranked open-weight model on the current Intelligence Index (2.8T MoE, 104B active). Scored: max effort	Moonshot / Kimi	Open weight	60	1.05M	$3	$15	Kimi K3 License (commercial with restrictions)	16 Jul 2026
GLM-5.3 Zhipu's current open-weight frontier; ties Kimi K3 at 60 on the Index. Scored: max effort	Zhipu / Z.AI	Open weight	60	1M	$1.40	$4.40	GLM-5 license	14 Aug 2026
Qwen3.8 2.4T A95B Alibaba's 2.4T MoE open-weight flagship (95B active, per the AA card).	Alibaba / Qwen	Open weight	58	984k	$2	$6	Qwen3 license	12 Aug 2026
GPT-5.6 Terra Mid GPT-5.6 tier: most of Sol's window at half the input price. Scored: max effort	OpenAI	Frontier closed	57	1.05M	$2	$12	—	9 Jul 2026
Muse Spark 1.2 Meta's API-only quality line; AA lists it proprietary, not Llama weights. Scored: xhigh	Meta	Frontier closed	57	1.05M	$1.25	$4.25	—	5 Aug 2026
Gemini 3.7 Flash Google's current high-throughput Gemini; scores above 3.1 Pro Preview on this Index. Scored: high	Google	Frontier closed	56	1M	$0.75	$3.75	—	13 Aug 2026
Claude Sonnet 5 Workhorse Claude: 1M context at Sonnet prices. Scored: adaptive reasoning, max effort	Anthropic	Frontier closed	55	1M	$2	$10	—	30 Jun 2026
DeepSeek V4 Pro 0813 1.6T MoE, 49B active, MIT weights; the current DeepSeek quality cut. Scored: max effort First-party API as measured by Artificial Analysis (may differ from off-peak list).	DeepSeek	Open weight	53	1M	$1.32	$3.96	MIT	13 Aug 2026
DeepSeek V4 Flash 0731 Cheaper DeepSeek V4 cut that still clears 50 on the Index. Scored: max effort First-party API as measured by Artificial Analysis.	DeepSeek	Open weight	52	1M	$0.44	$1.32	MIT	31 Jul 2026
GPT-5.6 Luna Volume GPT-5.6 tier: same 1.05M window, priced for high-throughput work. Scored: max effort	OpenAI	Frontier closed	52	1.05M	$0.20	$1.20	—	9 Jul 2026
Gemini 3.1 Pro Preview Google's Pro-class preview; Flash 3.7 now outscores it on this Index. AA lists cost per task, not a clean input/output pair we independently confirmed on Google's pricing page.	Google	Frontier closed	48	1M	—	—	—	19 Feb 2026
Claude Haiku 4.5 Fast, cheap Claude still on the current Anthropic price card (knowledge cutoff Feb 2025 on that page).	Anthropic	Frontier closed	30	200k	$1	$5	—	15 Oct 2025
Gemma 4 31B Google's current Gemma dense open-weight; Apache 2.0, 256k context. AA lists $0 / $0; treat as self-host, not a Google list price.	Google	Open weight	30	256k	—	—	Apache 2.0	2 Apr 2026
Mistral Medium 3.5 Mistral's current mid-size open-weight; Large 3 scores well below this on the Index.	Mistral	Open weight	30	256k	$1.50	$7.50	Modified MIT	29 Apr 2026
Llama 4 Maverick Still the Llama-branded open-weight you can download; far behind Muse Spark on this Index. Median provider price on AA, not a first-party Meta API.	Meta	Open weight	14	1M	$0.26	$0.91	Llama 4 Community License	5 Apr 2025

Sources

Figures as of 1 Sep 2026. They lag the labs. We fetched the pages below; we did not invent scores, windows, licenses, or list prices. Google's own Gemini 3.7 pricing page did not load in this pass, so Flash input/output dollars are Artificial Analysis's Google API listing, and Gemini 3.1 Pro Preview has no token price on this table.

Artificial Analysis LLM Leaderboard — Intelligence Index v4.1.1 scores, context windows, and first-party API input/output prices as published 1 Sep 2026.
Artificial Analysis Intelligence Index v4.1.1 — Composite of GDPval-AA v2, τ³-Banking, Terminal-Bench v2.1, SciCode, Humanity's Last Exam, GPQA Diamond, CritPt, AA-Omniscience, AA-LCR. Index patched 6 Aug 2026.
OpenAI — GPT-5.6 Sol — Official context (1,050,000), list price $4 / $20 per 1M tokens, promotional through at least 21 Nov 2026.
OpenAI — GPT-5.6 Terra — Official context and $2 / $12 per 1M tokens.
OpenAI — GPT-5.6 Luna — Official context and $0.20 / $1.20 per 1M tokens.
Anthropic — Claude Opus 5 overview — Released 24 Jul 2026. 1M context. Lineup prices: Fable $10/$50, Opus $5/$25, Sonnet $2/$10, Haiku 4.5 $1/$5.
Z.ai — GLM-5.3 — Announced 14 Aug 2026. Weights followed after a two-week safety pass.
xAI — Introducing Grok 4.6 — Released 12 Aug 2026.
xAI — Grok 4.6 docs — 500,000 context. $2 / $6 per 1M tokens on the standard band.
xAI API pricing — Confirms grok-4.6 flagship list price and 500k context.
Artificial Analysis — GPT-5.6 Sol (max) — Index 61; In $4 / Out $20; released 9 Jul 2026.
Artificial Analysis — Claude Opus 5 (max) — Index 63; In $5 / Out $25; released 24 Jul 2026.
Artificial Analysis — Claude Fable 5 — Index 62; In $10 / Out $50; released 9 Jun 2026.
Artificial Analysis — Claude Sonnet 5 (max) — Index 55; In $2 / Out $10; released 30 Jun 2026.
Artificial Analysis — Kimi K3 (max) — Index 60; open weights; Kimi K3 License (commercial with restrictions); In $3 / Out $15; released 16 Jul 2026.
Artificial Analysis — DeepSeek V4 Pro 0813 (max) — Index 53; MIT; In $1.32 / Out $3.96 on DeepSeek API as AA measured; released 13 Aug 2026.
Artificial Analysis — GLM-5.3 (max) — Index 60 on the leaderboard; GLM-5 license; In $1.40 / Out $4.40.
Artificial Analysis — Qwen3.8 2.4T A95B — Index 58; Qwen3 license; In $2 / Out $6; released 12 Aug 2026.
Artificial Analysis — Gemini 3.7 Flash (high) — Index 56; In $0.75 / Out $3.75 as AA lists Google API; August 2026.
Artificial Analysis — Grok 4.6 — Index 61 at high effort; In $2 / Out $6.
Artificial Analysis — Muse Spark 1.2 (xhigh) — Meta API model; Index 57; In $1.25 / Out $4.25; August 2026. AA labels it proprietary.
Artificial Analysis — Llama 4 Maverick — Open weights; Llama 4 Community License; Index 14; April 2025.
Artificial Analysis — Gemma 4 31B — Apache 2; Index 30; April 2026. AA lists $0/$0 (self-host / no first-party list price).
Artificial Analysis — Mistral Medium 3.5 — Modified MIT; Index 30; In $1.50 / Out $7.50; April 2026.
Artificial Analysis — DeepSeek V4 Flash 0731 (max) — MIT; Index 52; In $0.44 / Out $1.32 as AA measured; July 2026.