xAI / manual · models.dev

Grok 4.6

xAI's frontier model for long-running agents, coding, knowledge work, and visual projects

Encoding fertility 1.22×

Harnesses

Same model, different drivers. ChatGPT is not OpenCode. $ / MU is only set when every official task passed. $ / pass still shows on a partial run; it does not rank. Time is output tokens at 80 tok/s, not wall clock.

StackPassedAttemptsTime$ / MU$ / pass$ / hrTokens / passBurn$ / xAI
Grokhigh3/3417m 43s$0.000450$0.6447$5.21129,6121.14×$0.6447
Grokmedium3/3516m 32s$0.000385$0.5504$4.83114,0151.00×$0.5504

Official jobs

Pass or fail on each sealed job. Attempts include retries. Failed tries stay in the bill.

HarnessDurable work queue 2/2JSON Patch (RFC 6902 subset) 1/1Token bucket and sliding window 1/1
Grok · highpass · 2pass · 1pass · 1
Grok · mediumpass · 2——

Tokens by job

Billed tokens on each official job, including retries. Cache reads drop out of this count.

  • Grok · high

    230,145Durable work queue
    40,558JSON Patch (RFC 6902 subset)
    118,134Token bucket and sliding window
  • Grok · medium

    205,475Durable work queue
    —JSON Patch (RFC 6902 subset)
    —Token bucket and sliding window

Where the bill goes

Failed attempts stay in the bill. Cache hits are a discount, shown separately.

  • Grok · high388,837

    Cache hit 1,632,128 (discount, not extra work)

  • Grok · medium342,046

    Cache hit 1,299,456 (discount, not extra work)

  • Uncached in
  • Output
  • Thinking
  • Cache write

Retry tax

Attempts are harness calls, including failed hidden-verifier retries. Extra tries stay in the bill.

  • Grokhigh4 attempts · 1 extra / 3 jobs
  • Grokmedium5 attempts · 2 extra / 3 jobs

Cache hits

Hits divided by uncached input plus hits. No cache means the harness did not report a discount.

  • Grokhigh84%
  • Grokmedium83%

Time vs $ / MU

Decode time at 80 tok/s against what that finish cost. A long token bill is not the same as an expensive one.

0s17m 43s$0.000450

Encoding by slice

Native tokens per MU on the same frozen basket. Tool schemas are where fertility often doubles.

  • English prose30% of the basket0.97×
  • Code25% of the basket1.08×
  • Structured15% of the basket0.87×
  • Tool schemas10% of the basket1.11×
  • CJK10% of the basket3.33×
  • Instructions10% of the basket0.88×
SliceCharactersNative tokensFertilityTrue inSource
English prose4,4741,0880.97×$1.95official
Code2,3216291.08×$2.17official
Structured3,0676640.87×$1.73official
Tool schemas2,3786621.11×$2.23official
CJK8316913.33×$6.65official
Instructions1,1642550.88×$1.75official

Endpoints

ProviderSKUList inList outCache hitCache write
xAI
grok-4.6$2.00$6.00$0.50$2.00

All stacks