LLM Context Window Comparison

All seven models accept about 1M tokens: GPT-6.1 Sol leads at 1,050,000, Gemini 3.8 Flash and Kimi K3 accept 1,048,576, and MiniMax-M3.1-Flash is the cheapest 1,000,000-token option at $0.00210 per short request.

Updated · 18 sources · Methodology

NeedPickWhy
Context window (tokens)GPT-6.1 Sol1,050,000 tokens10
Lowest example costMiniMax-M3.1-Flash$0.002108
Fastest measured outputDeepSeek V4.1 Flash 1017169.0 tokens/s2

Capabilities and serving conditions

ModelOfficial capacity (tokens)AA reference context (tokens)Novita AI context / max output (tokens)Output speed (tokens/s)AA · HLE
GPT-6.1 Sol1,050,000101,000,0009Not on Novita AI54.7951.4%9
Gemini 3.8 Flash1,048,576131,000,00011Not on Novita AI132.21147.8%11
Kimi K31,048,576161,048,57641,048,576 / 1,048,5761852.7446.9%4
MiniMax-M3.1-Flash1,000,000171,000,00051,000,000 / 524,28818116.2548.0%5
DeepSeek V4.1 Flash 10171,000,000141,000,00021,048,576 / 393,21618169.0243.3%2
GLM-5.31,000,000151,000,00031,048,576 / 131,0721898.2342.3%3
Claude Opus 5.51,000,00061,000,0001Not on Novita AI96.8161.4%1

Prices and request cost

ModelInput / 1M tokensOutput / 1M tokensEstimated request cost
GPT-6.1 Sol10$2.00$10.00$0.03000
Gemini 3.8 Flash12$0.75$3.75$0.01125
Kimi K3718$3$15$0.04500
MiniMax-M3.1-Flash818$0.15$0.60$0.00210
DeepSeek V4.1 Flash 1017718$0.30$1.20$0.00420
GLM-5.3718$1.40$4.40$0.01840
Claude Opus 5.56$4$20$0.06000
Cost example: 10,000 input + 1,000 billed output tokens
GPT-6.1 Sol$0.03000
Gemini 3.8 Flash$0.01125
Kimi K3$0.04500
MiniMax-M3.1-Flash$0.00210
DeepSeek V4.1 Flash 1017$0.00420
GLM-5.3$0.01840
Claude Opus 5.5$0.06000

Options, strengths and limitations

  1. GPT-6.1 Sol

    The largest published window here, 1,050,000 tokens, 1,424 more than Gemini 3.8 Flash and Kimi K3.109

    Context window (tokens)
    1,050,000
    Output speed (tokens/s)
    54.7
    Request cost ($)
    0.03000

    Strengths

    Its 51.4% HLE score is second only to Claude Opus 5.5’s 61.4%, at half Claude’s $0.06000 short-request cost.109

    Limitations

    Once input exceeds 272,000 tokens the whole request is billed at 2× input and 1.5× output, so a 500,000-token prompt costs $2.00 in input alone.109

    109

  2. Gemini 3.8 Flash

    1,048,576 tokens at the lowest proprietary-model price here, $0.01125 per short request.131112

    Context window (tokens)
    1,048,576
    Output speed (tokens/s)
    132.2
    Request cost ($)
    0.01125

    Strengths

    Its 132.2 tokens/s is the second-fastest output here, behind DeepSeek V4.1 Flash 1017’s 169.0.131112

    Limitations

    The text rates are promotional through 2026-12-31; from 2027-01-01 they double to $1.50 input and $7.50 output, and audio input is billed separately.131112

    131112

  3. Kimi K3

    1,048,576 tokens, and on Novita AI the same 1,048,576-token limit applies to output.164718

    Context window (tokens)
    1,048,576
    Output speed (tokens/s)
    52.7
    Request cost ($)
    0.04500

    Strengths

    Its 1,048,576-token output limit on Novita AI is 2.7× DeepSeek V4.1 Flash 1017’s 393,216 and 8× GLM-5.3’s 131,072 on the same platform.164718

    Limitations

    At 52.7 tokens/s it is the slowest model here, and its $0.04500 short request is the second most expensive.164718

    164718

  4. MiniMax-M3.1-Flash

    1,000,000 tokens at the lowest short-request cost here, $0.00210.175818

    Context window (tokens)
    1,000,000
    Output speed (tokens/s)
    116.2
    Request cost ($)
    0.00210

    Strengths

    Its 48.0% HLE score is the highest of the four models Novita AI serves here, and Novita AI allows up to 524,288 output tokens.175818

    Limitations

    Its 116.2 tokens/s is about 31% below DeepSeek V4.1 Flash 1017’s 169.0, and its window is 50,000 tokens below GPT-6.1 Sol’s.175818

    175818

  5. DeepSeek V4.1 Flash 1017

    The fastest output here, 169.0 tokens/s, with a 1,000,000-token published window.142718

    Context window (tokens)
    1,000,000
    Output speed (tokens/s)
    169.0
    Request cost ($)
    0.00420

    Strengths

    At $0.00420 per short request it is the second-cheapest model here, and Novita AI serves it with 1,048,576 context and 393,216 output tokens.142718

    Limitations

    Its 43.3% HLE score is 18.1 points below Claude Opus 5.5’s 61.4%.142718

    142718

  6. GLM-5.3

    1,000,000 published tokens; Novita AI serves 1,048,576 context with a 131,072-token output cap.153718

    Context window (tokens)
    1,000,000
    Output speed (tokens/s)
    98.2
    Request cost ($)
    0.01840

    Strengths

    Its $0.01840 short request costs less than GPT-6.1 Sol’s $0.03000 and Kimi K3’s $0.04500.153718

    Limitations

    Its 42.3% HLE score is the lowest here, and its 131,072-token output cap is the smallest of the four models on Novita AI.153718

    153718

  7. Claude Opus 5.5

    The top HLE score here, 61.4%, with a 1,000,000-token window.61

    Context window (tokens)
    1,000,000
    Output speed (tokens/s)
    96.8
    Request cost ($)
    0.06000

    Strengths

    Its 61.4% HLE leads GPT-6.1 Sol by 10.0 points, at 96.8 tokens/s.61

    Limitations

    Its $0.06000 short request is the most expensive here, 29× MiniMax-M3.1-Flash’s $0.00210.61

    61

Methodology and sources

Published capacity uses each vendor’s model documentation or model card. AA’s tested context, HLE and output speed come from AA model pages, at max reasoning for Claude Opus 5.5 (default fallback), DeepSeek V4.1 Flash 1017, GLM-5.3 and Kimi K3, and high reasoning for GPT-6.1 Sol and Gemini 3.8 Flash. Novita AI context and output limits come from its pricing page. The short cost example is 10,000 input and 1,000 billed output tokens, not a full-window prompt or a retrieval benchmark.

Costs are estimates at listed rates for uncached synchronous requests, not measured invoices; billed output includes reasoning. Tax, retries, discounts and negotiated capacity are excluded. Novita AI publishes this page using public sources, without a private workload benchmark.

OpenAI lists GPT-6.1 Sol at 1,050,000 tokens and Google lists Gemini 3.8 Flash at 1,048,576; AA rounds both to 1,000,000. The DeepSeek V4.1 Flash 1017 and GLM-5.3 model cards state 1,000,000 tokens, while Novita AI’s endpoints for both accept 1,048,576.1013141518

This context ranking orders cards by officially published token capacity, then by the short uncached request cost for equal capacities. Official capacity and AA-tested context are separate columns; capacity is not a retrieval-accuracy result.

  1. Artificial Analysis · Claude Opus 5.5 (Max, Default Fallback)
  2. Artificial Analysis · DeepSeek V4.1 Flash 1017 (Max)
  3. Artificial Analysis · GLM-5.3 (Max)
  4. Artificial Analysis · Kimi K3 (Max)
  5. Artificial Analysis · MiniMax-M3.1-Flash
  6. Anthropic · Opus 5.5 pricing
  7. Together AI · model API prices
  8. MiniMax · pay-as-you-go pricing
  9. Artificial Analysis · GPT-6.1 Sol (High)
  10. OpenAI · GPT-6.1 Sol
  11. Artificial Analysis · Gemini 3.8 Flash (High)
  12. Google · Gemini API
  13. Google · Gemini 3.8 Flash
  14. DeepSeek · official model context
  15. GLM · official model context
  16. Kimi · official model context
  17. MiniMax · official model context
  18. Novita AI · Pricing

Build with Novita Model APIs

Create an API key and connect your application.

Start BuildingExplore Models