Open models, explained.

Which open model is best for your use case, what it costs, and how it compares to closed models.

Pick by use case

One index hides the fit. Choose what you are building; the answer and the evidence follow.

Coding Agents Knowledge Long context Multilingual VisionEngineering Legal Healthcare Finance Economics Strategy

Which model should power our coding agent?

42 models scored · 3 measures · Artificial Analysis

Open pick

Kimi K376

$0.84 per task · 93% of the leader's score for 72% less

Best value

MiMiMo-V2.557

$0.01 per task · 75% of the open pick, 1% of the cost

To beat

Claude Fable 5.182

$3.69 per task · 5.4 points ahead of the open pick

Open sourceClosed

Score breakdownAA Coding Index · Terminal-Bench 2.1 · SciCode

Model	Score	Cost / task	AA Coding Index	Terminal-Bench 2.1	SciCode
Claude Fable 5.1	81.6	$3.69	81.6	91.4	62.0
Claude Opus 5	78.0	$2.34	78.0	89.1	55.7
GPT-5.6 Sol	77.4	$0.95	77.4	88.0	56.1
GPT-6 Astra	76.9	$1.67	76.9	88.4	54.1
Grok 4.6	76.8	$0.94	76.8	88.4	53.6
GPT-5.6 Terra	76.7	$0.53	76.7	88.0	53.9
Claude Fable 5	76.5	$3.14	76.5	84.6	60.2
Gemini 3.8 Flash	76.3	$0.58	76.3	87.6	53.6
Kimi K3	76.2	$0.84	76.2	85.0	58.7
Gemini 3.7 Flash	76.1	$0.40	76.1	85.8	56.8

Best result per column is underlined. A “·” means Artificial Analysis has not run that measure against the model; it is left as a gap, never counted as a zero.

Every model, ranked

The whole field on one index. Sort, filter, click to inspect.

Artificial Analysis Intelligence Index

Composite of nine public evals: agents, coding, reasoning, knowledge · Higher is better

Open sourceClosed

Click a bar to inspect

top 18 of 42 shown · full list belowAttributedArtificial Analysis v4.1.1

Leaderboard

	Model							License
1	Claude Fable 5.1 Anthropic	66	0.8	$8.5K	$10.0 / $50.0	67 t/s	1M	Proprietary
2	Claude Opus 5 Anthropic	63	1.6	$3.8K	$5.00 / $25.0	58 t/s	1M	Proprietary
3	Claude Fable 5 Anthropic	62	1.1	$5.5K	$10.0 / $50.0	70 t/s	1M	Proprietary
4	GPT-6 Astra OpenAI	61	2.0	$3.0K	$10.0 / $50.0	·	1M	Proprietary
5	GPT-5.6 Sol OpenAI	61	3.0	$2.0K	$4.00 / $20.0	81 t/s	1M	Proprietary
6	Grok 4.6 xAI	61	5.3	$1.2K	$2.00 / $6.00	65 t/s	500K	Proprietary
7	Kimi K3 Moonshot (Kimi)	60	2.5	$2.4K	$3.00 / $15.0	62 t/s	1.0M	Kimi K3 License
8	Z GLM-5.3 Z.ai (GLM)	60	4.8	$1.2K	$1.40 / $4.40	134 t/s	1M	GLM-5.3 License
9	Gemini 3.8 Flash Google	59	7.1	$826	$0.75 / $3.75	306 t/s	1M	Proprietary
10	Qwen3.8 Max Alibaba (Qwen)	58	3.8	$1.5K	$2.00 / $6.00	39 t/s	1M	Proprietary

Kimi K3

Moonshot (Kimi) · Kimi K3 License · 2800B · 104B active

open

Intelligence Index60

Intel / $1002.5

Cost per task: $0.84
Full index run: $2.4K
Price in / out: $3.00 / $15.0
Cached input: 10x cheaper$0.30
Speed: 1.01s TTFT62 t/s
Context: 1.0M

Capability breakdown0 to 100

Coding

Agents

GPQA

HLE

Long context

moonshotai/Kimi-K3Run it

The state of open, at a glance

Who is smartest, and who gives the most per dollar.

Open sourceClosed

Intelligence

Artificial Analysis Intelligence Index · Higher is better

Value

Index points per $100 of run cost · models scoring 50+

Kimi K3 is 6 points behind Claude Fable 5.1 on the Intelligence Index, holding 91% of the top score. In 2025 that gap was 11.2 points, and it has narrowed since.

Kimi K3

Open source · 60

Claude Fable 5.1

Closed · 66

Run the open frontier yourself

Same OpenAI-compatible API, open source weights you keep, at a fraction of closed-API cost. The snippet is pointed at Kimi K3, the highest-scoring open model Together serves today and the open leader overall.

Start on Together What would it cost?

Copied

# uv add together

from together import Together

client = Together()

response = client.chat.completions.create(

model="moonshotai/Kimi-K3",

messages=[{"role": "user", "content": "What are the top 3 things to do in New York?"}],

)

print(response.choices[0].message.content)