GPT-6 Astra vs Kimi K3

At a glance

• Capability index

  • GPT-6 Astra166.4
  • Kimi K3157.4

  GPT-6 Astra is 9.0 points higher.

• Price per 1M tokens

  • GPT-6 Astra$20.00
  • Kimi K3$6.00

  Kimi K3 is 3.3 times cheaper.

• Context window

  • GPT-6 Astra1.1M tokens
  • Kimi K31M tokens

  GPT-6 Astra takes in 0% more.

• Released

  • GPT-6 AstraSep 4, 2026
  • Kimi K3Jul 16, 2026

  GPT-6 Astra is 50 days newer.

Where each scores higher

On the 22 tests both have taken, each on its own scale. A gap of a point or two can sit within a test’s margin of error.

Everything side by side

Comparison of GPT-6 Astra vs Kimi K3

Measure	GPT-6 AstraOpenAI	Kimi K3Moonshot AI
Model
Maker	OpenAI	Moonshot AI
Sold here by	OpenAI	Moonshot AI
API model ID	gpt-6-astra	kimi-k3
Released	Sep 4, 2026	Jul 16, 2026
Knowledge cutoff	Apr 30, 2026	Not reported
Weights	Closed: hosted access only	Open: downloadable
Developer’s country	United States	China
Hosts selling it	29including OpenAI directly	84including Moonshot AI directly
Training compute	1.0 × 10²⁷ FLOPEpoch AI estimate	2.0 × 10²⁵ FLOPEpoch AI estimate
Test results
Capability index	166.4 (best of these models)#2 of 274	157.4#15 of 274
LiveBench overall	82.2% (best of these models)#4 of 66	79.2%#15 of 66
Arena text	1477#29 of 413	1488 (best of these models)#16 of 413
LiveBench coding	80.4%#20 of 66	81.5% (best of these models)#15 of 66
Arena WebDev	1788 (best of these models)#2 of 138	1658#13 of 138
Arena coding	1543 (best of these models)#7 of 408	1541#9 of 408
LiveBench mathematics	96.8% (best of these models)#4 of 66	84.4%#53 of 66
Mock AIME 2024–2025	100.0% (best of these models)#1 of 297	97.2%#31 of 297
FrontierMath Tiers 1–3	93.7% (best of these models)#1 of 114	72.2%#22 of 114
FrontierMath Tier 4	97.6% (best of these models)#2 of 70	39.0%#29 of 70
Arena math	1490#27 of 396	1501 (best of these models)#16 of 396
SimpleQA Verified	75.6% (best of these models)#1 of 86	50.6%#24 of 86
LiveBench agentic coding	57.3%#21 of 66	62.2% (best of these models)#10 of 66
DeepSWE	74.1% (best of these models)#1 of 69	68.5%#19 of 69
LiveBench language	89.4% (best of these models)#4 of 66	85.5%#11 of 66
LiveBench instructions	75.6% (best of these models)#8 of 66	71.4%#23 of 66
Arena creative writing	1449#39 of 411	1460 (best of these models)#26 of 411
Arena instruction following	1474#30 of 413	1488 (best of these models)#14 of 413
LiveBench data analysis	83.0% (best of these models)#1 of 66	78.7%#21 of 66
LiveBench reasoning	92.7% (best of these models)#1 of 66	90.7%#9 of 66
GPQA Diamond	95.8% (best of these models)#1 of 319	93.1%#20 of 319
Chess puzzles	72.0% (best of these models)#1 of 227	39.0%#31 of 227
Arena hard prompts	1502#30 of 413	1517 (best of these models)#9 of 413
Price per million tokens
Input	$10.00	$3.00 (best of these models)
Output	$50.00	$15.00 (best of these models)
Blended, 3 input to 1 output	$20.00	$6.00 (best of these models)
Cached input	$1.00	$0.30 (best of these models)
Writing to the cache	$12.50	Not reported
Long requests	$20.00 in, $75.00 outabove 272K tokens	No separate price reported
The maker’s own price	This listing	This listing
Limits
Context window	1.1M tokens (best of these models)	1M tokens
Max input	922K tokens	1M tokens (best of these models)
Max output	128K tokens	1M tokens (best of these models)
Features
Reads	images, PDFs and text	images, text and video
Produces	text	text
Reasoning	Yeseffort low, medium, high, xhigh, max	Yeseffort low, high, max; can be switched off; thinking budget can be set
Tool calling	Yes	Yes
Structured output	Yes, such as JSON	Yes, such as JSON
File attachments	Yes	Yes
Temperature setting	Fixed by the provider	Fixed by the provider

Bold marks the better value in each row. Prices are each model’s own list price, or the middle price across hosts when the maker does not sell it, blended as three input tokens for every output token. Test results as published by LiveBench (Apache-2.0), Epoch AI (CC BY 4.0), Datacurve DeepSWE, compiled by Epoch AI (CC BY 4.0), Arena (CC BY 4.0); model details and prices from models.dev (MIT). Confirm pricing with the provider before use.