geekradar
Int
Explore GeekRadar

More to explore

Good gear All topics About us Say hello
GeekRadar / The comparison desk
COMPARE.
AI models
minimax

MiniMax-M3

AI model Released

API mode
Chat*
Context window
1,048,576 tokens
Maximum output
512,000 tokens
Open weights
Yes

Capability & preference

General capability · ECI

A cross-benchmark capability estimate. Index points, not a percentage; intervals show uncertainty.

MiniMax-M3 146.9* MiniMax-M3 · 95% interval 142.4–150.3

Human preference · Arena

Overall text preferences from head-to-head votes. Not factual accuracy or a task-success rate.

MiniMax-M3 1431.3 MiniMax-M3 · 95% interval 1427.5–1435.0 · 68,611 votes · 2026-10-28

Model performance

Radar task score: four tests, equal weight. An experimental summary of these tasks, not a universal intelligence rating.

MiniMax-M3

Radar task score

Unrated

2/4 core tests · not enough evidence

3 benchmark results for this configuration

Scientific reasoning

GPQA Diamond · 25%

Expert-written science questions; strict multiple-choice scoring. Chance accuracy is 25%.

MiniMax-M3 90.9*% Approx. 95% interval: 86.9–94.9%

Novel problem solving

Mystery Game Puzzles · 25%

Best-next-move puzzles in an undisclosed game variant. A narrow reasoning test, not general intelligence.

MiniMax-M3 7.0*% Approx. 95% interval: 2.0–12.0%

Advanced mathematics

FrontierMath Tiers 1-3 · 25%

Private v2 mathematics problems with Python tools. Earlier versions and Tier 4 are excluded.

MiniMax-M3 Not available

Factual answers

SimpleQA Verified · 25%

Correct factual answers under a guessing prompt, graded by a model. Not a hallucination or abstention score.

MiniMax-M3 Not available

Competition mathematics

OTIS Mock AIME 2024-2025 · Separate result

Competition problems with model-extracted answers checked against an exact key. Earlier grading procedures are excluded.

MiniMax-M3 71.1*% Approx. 95% interval: 57.7–84.5%

Bars show reported task scores. Whiskers are approximate 95% intervals where standard errors are available. Different reasoning budgets are not equal-cost tests. Missing results are never zero.

Recorded results

Only the same benchmark, version, workload, scenario and unit share a row. Missing results are not zero. System results include the full submitted hardware and software configuration.

Arena text preference / chinesehuman preference · 2026-10-28 · rating

MiniMax-M3

1,464.99 rating

Confidence interval: 1,455.38 – 1,474.59

4461 votes

Category
chinese
Evaluated configuration
MiniMax-M3
Evaluation type
human preference
Arena text preference / codinghuman preference · 2026-10-28 · rating

MiniMax-M3

1,469.50 rating

Confidence interval: 1,463.97 – 1,475.03

19379 votes

Category
coding
Evaluated configuration
MiniMax-M3
Evaluation type
human preference
Arena text preference / creative_writinghuman preference · 2026-10-28 · rating

MiniMax-M3

1,400.45 rating

Confidence interval: 1,394.06 – 1,406.85

12662 votes

Category
creative_writing
Evaluated configuration
MiniMax-M3
Evaluation type
human preference
Arena text preference / englishhuman preference · 2026-10-28 · rating

MiniMax-M3

1,441.54 rating

Confidence interval: 1,436.70 – 1,446.39

28097 votes

Category
english
Evaluated configuration
MiniMax-M3
Evaluation type
human preference
Arena text preference / exclude_tieshuman preference · 2026-10-28 · rating

MiniMax-M3

1,423.09 rating

Confidence interval: 1,418.23 – 1,427.94

51553 votes

Category
exclude_ties
Evaluated configuration
MiniMax-M3
Evaluation type
human preference
Arena text preference / experthuman preference · 2026-10-28 · rating

MiniMax-M3

1,461.29 rating

Confidence interval: 1,453.69 – 1,468.89

7463 votes

Category
expert
Evaluated configuration
MiniMax-M3
Evaluation type
human preference
Arena text preference / frenchhuman preference · 2026-10-28 · rating

MiniMax-M3

1,440.75 rating

Confidence interval: 1,427.03 – 1,454.46

2502 votes

Category
french
Evaluated configuration
MiniMax-M3
Evaluation type
human preference
Arena text preference / germanhuman preference · 2026-10-28 · rating

MiniMax-M3

1,428.63 rating

Confidence interval: 1,410.28 – 1,446.97

1181 votes

Category
german
Evaluated configuration
MiniMax-M3
Evaluation type
human preference
Arena text preference / hard_promptshuman preference · 2026-10-28 · rating

MiniMax-M3

1,446.01 rating

Confidence interval: 1,441.65 – 1,450.36

45352 votes

Category
hard_prompts
Evaluated configuration
MiniMax-M3
Evaluation type
human preference
Arena text preference / hard_prompts_englishhuman preference · 2026-10-28 · rating

MiniMax-M3

1,453.88 rating

Confidence interval: 1,448.38 – 1,459.39

19169 votes

Category
hard_prompts_english
Evaluated configuration
MiniMax-M3
Evaluation type
human preference
Arena text preference / industry_business_and_management_and_financial_operationshuman preference · 2026-10-28 · rating

MiniMax-M3

1,433.05 rating

Confidence interval: 1,426.87 – 1,439.23

13167 votes

Category
industry_business_and_management_and_financial_operations
Evaluated configuration
MiniMax-M3
Evaluation type
human preference
Arena text preference / industry_entertainment_and_sports_and_mediahuman preference · 2026-10-28 · rating

MiniMax-M3

1,400.95 rating

Confidence interval: 1,395.07 – 1,406.84

16131 votes

Category
industry_entertainment_and_sports_and_media
Evaluated configuration
MiniMax-M3
Evaluation type
human preference
Arena text preference / industry_legal_and_governmenthuman preference · 2026-10-28 · rating

MiniMax-M3

1,434.03 rating

Confidence interval: 1,425.31 – 1,442.74

5555 votes

Category
industry_legal_and_government
Evaluated configuration
MiniMax-M3
Evaluation type
human preference
Arena text preference / industry_life_and_physical_and_social_sciencehuman preference · 2026-10-28 · rating

MiniMax-M3

1,449.68 rating

Confidence interval: 1,443.20 – 1,456.16

10896 votes

Category
industry_life_and_physical_and_social_science
Evaluated configuration
MiniMax-M3
Evaluation type
human preference
Arena text preference / industry_mathematicalhuman preference · 2026-10-28 · rating

MiniMax-M3

1,446.54 rating

Confidence interval: 1,436.44 – 1,456.64

3964 votes

Category
industry_mathematical
Evaluated configuration
MiniMax-M3
Evaluation type
human preference
Arena text preference / industry_medicine_and_healthcarehuman preference · 2026-10-28 · rating

MiniMax-M3

1,437.76 rating

Confidence interval: 1,428.33 – 1,447.19

4616 votes

Category
industry_medicine_and_healthcare
Evaluated configuration
MiniMax-M3
Evaluation type
human preference
Arena text preference / industry_software_and_it_serviceshuman preference · 2026-10-28 · rating

MiniMax-M3

1,465.92 rating

Confidence interval: 1,460.97 – 1,470.86

27474 votes

Category
industry_software_and_it_services
Evaluated configuration
MiniMax-M3
Evaluation type
human preference
Arena text preference / industry_writing_and_literature_and_languagehuman preference · 2026-10-28 · rating

MiniMax-M3

1,416.24 rating

Confidence interval: 1,410.57 – 1,421.90

17551 votes

Category
industry_writing_and_literature_and_language
Evaluated configuration
MiniMax-M3
Evaluation type
human preference
Arena text preference / instruction_followinghuman preference · 2026-10-28 · rating

MiniMax-M3

1,431.04 rating

Confidence interval: 1,425.91 – 1,436.17

23843 votes

Category
instruction_following
Evaluated configuration
MiniMax-M3
Evaluation type
human preference
Arena text preference / japanesehuman preference · 2026-10-28 · rating

MiniMax-M3

1,375.51 rating

Confidence interval: 1,355.17 – 1,395.85

1064 votes

Category
japanese
Evaluated configuration
MiniMax-M3
Evaluation type
human preference
Arena text preference / koreanhuman preference · 2026-10-28 · rating

MiniMax-M3

1,370.68 rating

Confidence interval: 1,352.37 – 1,388.99

1241 votes

Category
korean
Evaluated configuration
MiniMax-M3
Evaluation type
human preference
Arena text preference / longer_queryhuman preference · 2026-10-28 · rating

MiniMax-M3

1,443.54 rating

Confidence interval: 1,438.60 – 1,448.49

30758 votes

Category
longer_query
Evaluated configuration
MiniMax-M3
Evaluation type
human preference
Arena text preference / mathhuman preference · 2026-10-28 · rating

MiniMax-M3

1,427.55 rating

Confidence interval: 1,417.02 – 1,438.08

3455 votes

Category
math
Evaluated configuration
MiniMax-M3
Evaluation type
human preference
Arena text preference / multi_turnhuman preference · 2026-10-28 · rating

MiniMax-M3

1,441.39 rating

Confidence interval: 1,434.95 – 1,447.83

11921 votes

Category
multi_turn
Evaluated configuration
MiniMax-M3
Evaluation type
human preference
Arena text preference / non_englishhuman preference · 2026-10-28 · rating

MiniMax-M3

1,418.33 rating

Confidence interval: 1,413.94 – 1,422.73

40507 votes

Category
non_english
Evaluated configuration
MiniMax-M3
Evaluation type
human preference
Arena text preference / overallhuman preference · 2026-10-28 · rating

MiniMax-M3

1,431.27 rating

Confidence interval: 1,427.55 – 1,434.99

68611 votes

Category
overall
Evaluated configuration
MiniMax-M3
Evaluation type
human preference
Arena text preference / polishhuman preference · 2026-10-28 · rating

MiniMax-M3

1,423.51 rating

Confidence interval: 1,406.33 – 1,440.68

1195 votes

Category
polish
Evaluated configuration
MiniMax-M3
Evaluation type
human preference
Arena text preference / russianhuman preference · 2026-10-28 · rating

MiniMax-M3

1,426.20 rating

Confidence interval: 1,418.92 – 1,433.48

8155 votes

Category
russian
Evaluated configuration
MiniMax-M3
Evaluation type
human preference
Arena text preference / spanishhuman preference · 2026-10-28 · rating

MiniMax-M3

1,429.90 rating

Confidence interval: 1,415.85 – 1,443.95

2128 votes

Category
spanish
Evaluated configuration
MiniMax-M3
Evaluation type
human preference
Arena text preference / chinesehuman preference · 2026-10-15 · rating

MiniMax-M3

1,467.41 rating

Confidence interval: 1,457.43 – 1,477.39

4069 votes

Category
chinese
Evaluated configuration
minimax-m3
Evaluation type
human preference
Arena text preference / codinghuman preference · 2026-10-15 · rating

MiniMax-M3

1,471.34 rating

Confidence interval: 1,465.70 – 1,476.98

18242 votes

Category
coding
Evaluated configuration
minimax-m3
Evaluation type
human preference
Arena text preference / creative_writinghuman preference · 2026-10-15 · rating

MiniMax-M3

1,404.30 rating

Confidence interval: 1,397.73 – 1,410.86

11713 votes

Category
creative_writing
Evaluated configuration
minimax-m3
Evaluation type
human preference
Arena text preference / englishhuman preference · 2026-10-15 · rating

MiniMax-M3

1,443.80 rating

Confidence interval: 1,438.84 – 1,448.75

26416 votes

Category
english
Evaluated configuration
minimax-m3
Evaluation type
human preference
Arena text preference / exclude_tieshuman preference · 2026-10-15 · rating

MiniMax-M3

1,425.80 rating

Confidence interval: 1,420.82 – 1,430.77

48119 votes

Category
exclude_ties
Evaluated configuration
minimax-m3
Evaluation type
human preference
Arena text preference / experthuman preference · 2026-10-15 · rating

MiniMax-M3

1,462.56 rating

Confidence interval: 1,454.75 – 1,470.37

6992 votes

Category
expert
Evaluated configuration
minimax-m3
Evaluation type
human preference
Arena text preference / frenchhuman preference · 2026-10-15 · rating

MiniMax-M3

1,444.80 rating

Confidence interval: 1,430.61 – 1,458.99

2295 votes

Category
french
Evaluated configuration
minimax-m3
Evaluation type
human preference
Arena text preference / germanhuman preference · 2026-10-15 · rating

MiniMax-M3

1,427.17 rating

Confidence interval: 1,408.35 – 1,445.98

1104 votes

Category
german
Evaluated configuration
minimax-m3
Evaluation type
human preference
Arena text preference / hard_promptshuman preference · 2026-10-15 · rating

MiniMax-M3

1,447.90 rating

Confidence interval: 1,443.44 – 1,452.37

42422 votes

Category
hard_prompts
Evaluated configuration
minimax-m3
Evaluation type
human preference
Arena text preference / hard_prompts_englishhuman preference · 2026-10-15 · rating

MiniMax-M3

1,455.58 rating

Confidence interval: 1,449.95 – 1,461.21

18062 votes

Category
hard_prompts_english
Evaluated configuration
minimax-m3
Evaluation type
human preference
Arena text preference / industry_business_and_management_and_financial_operationshuman preference · 2026-10-15 · rating

MiniMax-M3

1,435.28 rating

Confidence interval: 1,428.94 – 1,441.62

12339 votes

Category
industry_business_and_management_and_financial_operations
Evaluated configuration
minimax-m3
Evaluation type
human preference
Arena text preference / industry_entertainment_and_sports_and_mediahuman preference · 2026-10-15 · rating

MiniMax-M3

1,402.85 rating

Confidence interval: 1,396.82 – 1,408.87

15034 votes

Category
industry_entertainment_and_sports_and_media
Evaluated configuration
minimax-m3
Evaluation type
human preference
Arena text preference / industry_legal_and_governmenthuman preference · 2026-10-15 · rating

MiniMax-M3

1,437.75 rating

Confidence interval: 1,428.80 – 1,446.70

5214 votes

Category
industry_legal_and_government
Evaluated configuration
minimax-m3
Evaluation type
human preference
Arena text preference / industry_life_and_physical_and_social_sciencehuman preference · 2026-10-15 · rating

MiniMax-M3

1,452.90 rating

Confidence interval: 1,446.26 – 1,459.54

10165 votes

Category
industry_life_and_physical_and_social_science
Evaluated configuration
minimax-m3
Evaluation type
human preference
Arena text preference / industry_mathematicalhuman preference · 2026-10-15 · rating

MiniMax-M3

1,447.36 rating

Confidence interval: 1,436.86 – 1,457.85

3630 votes

Category
industry_mathematical
Evaluated configuration
minimax-m3
Evaluation type
human preference
Arena text preference / industry_medicine_and_healthcarehuman preference · 2026-10-15 · rating

MiniMax-M3

1,442.18 rating

Confidence interval: 1,432.46 – 1,451.90

4321 votes

Category
industry_medicine_and_healthcare
Evaluated configuration
minimax-m3
Evaluation type
human preference
Arena text preference / industry_software_and_it_serviceshuman preference · 2026-10-15 · rating

MiniMax-M3

1,467.35 rating

Confidence interval: 1,462.30 – 1,472.40

25769 votes

Category
industry_software_and_it_services
Evaluated configuration
minimax-m3
Evaluation type
human preference
Arena text preference / industry_writing_and_literature_and_languagehuman preference · 2026-10-15 · rating

MiniMax-M3

1,419.68 rating

Confidence interval: 1,413.87 – 1,425.49

16321 votes

Category
industry_writing_and_literature_and_language
Evaluated configuration
minimax-m3
Evaluation type
human preference
Arena text preference / instruction_followinghuman preference · 2026-10-15 · rating

MiniMax-M3

1,433.18 rating

Confidence interval: 1,427.92 – 1,438.43

22259 votes

Category
instruction_following
Evaluated configuration
minimax-m3
Evaluation type
human preference
Arena text preference / japanesehuman preference · 2026-10-15 · rating

MiniMax-M3

1,380.61 rating

Confidence interval: 1,359.17 – 1,402.06

953 votes

Category
japanese
Evaluated configuration
minimax-m3
Evaluation type
human preference
Arena text preference / koreanhuman preference · 2026-10-15 · rating

MiniMax-M3

1,373.28 rating

Confidence interval: 1,354.68 – 1,391.89

1192 votes

Category
korean
Evaluated configuration
minimax-m3
Evaluation type
human preference
Arena text preference / longer_queryhuman preference · 2026-10-15 · rating

MiniMax-M3

1,445.94 rating

Confidence interval: 1,440.88 – 1,451.00

28906 votes

Category
longer_query
Evaluated configuration
minimax-m3
Evaluation type
human preference
Arena text preference / mathhuman preference · 2026-10-15 · rating

MiniMax-M3

1,429.32 rating

Confidence interval: 1,418.43 – 1,440.21

3192 votes

Category
math
Evaluated configuration
minimax-m3
Evaluation type
human preference
Arena text preference / multi_turnhuman preference · 2026-10-15 · rating

MiniMax-M3

1,442.97 rating

Confidence interval: 1,436.35 – 1,449.59

11126 votes

Category
multi_turn
Evaluated configuration
minimax-m3
Evaluation type
human preference
Arena text preference / non_englishhuman preference · 2026-10-15 · rating

MiniMax-M3

1,420.17 rating

Confidence interval: 1,415.66 – 1,424.68

37548 votes

Category
non_english
Evaluated configuration
minimax-m3
Evaluation type
human preference
Arena text preference / overallhuman preference · 2026-10-15 · rating

MiniMax-M3

1,433.33 rating

Confidence interval: 1,429.51 – 1,437.14

63973 votes

Category
overall
Evaluated configuration
minimax-m3
Evaluation type
human preference
Arena text preference / polishhuman preference · 2026-10-15 · rating

MiniMax-M3

1,423.58 rating

Confidence interval: 1,405.80 – 1,441.35

1112 votes

Category
polish
Evaluated configuration
minimax-m3
Evaluation type
human preference
Arena text preference / russianhuman preference · 2026-10-15 · rating

MiniMax-M3

1,426.72 rating

Confidence interval: 1,419.19 – 1,434.25

7506 votes

Category
russian
Evaluated configuration
minimax-m3
Evaluation type
human preference
Arena text preference / spanishhuman preference · 2026-10-15 · rating

MiniMax-M3

1,432.38 rating

Confidence interval: 1,417.81 – 1,446.95

1959 votes

Category
spanish
Evaluated configuration
minimax-m3
Evaluation type
human preference

Run it locally

Inference · memory estimate

MiniMax-M3

795.5 GiB checkpoint tensor payload · 427.04 B stored tensor elements · BF16/F32

Stored tensors can include buffers and repeated shared weights. A verified unique parameter count is unavailable. Checkpoint size is not the memory required to run the model.

API cost

USD / month

Standard text-token estimate. Excludes taxes, caching, batch discounts, reasoning and multimodal surcharges. Provider limits and regional availability may differ.

MiniMax-M3

MiniMax-M3 / Pricing tiers 2
Provider / scopeRateCurrencyAmount / unit
DeepInfraMiniMaxAI/MiniMax-M3Standard Cached inputUSD0.056 / 1M tokens
InputUSD0.28 / 1M tokens
OutputUSD1.1 / 1M tokens
SambaNova CloudMiniMax-M3Standard InputUSD0.6 / 1M tokens
OutputUSD2.4 / 1M tokens

Capabilities

Context window
1,048,576 tokens
Maximum input
1,000,000 tokens*
Maximum output
512,000 tokens
Reasoning
Yes
Tool calling
Yes
Open weights
Yes
API mode
Chat*
Image input
Yes*
Prompt caching
Yes*
System messages
Yes*
Tool choice
Yes*
Checkpoint tensor payload
795.51 GiB

Input modalities

  • text
  • image
  • video

Output modalities

  • text

Processing & platform

Architecture
minimax_m3_vl

System configuration

Detailed specifications

Specifications · 16 fields
cache read input token cost
6.0e-8*
Scope and conditions
native_context
provider
minimax*
mode
chat*
full_native_model_id
minimax/MiniMax-M3*
pricing_scope
Literal source field names and values retained; provider, region, cache/batch/threshold conditions and per-unit meaning must be checked. max_tokens is not automatically a context limit.*
cache read input token cost above 512k tokens
1.2e-7*
Scope and conditions
native_context
provider
minimax*
mode
chat*
full_native_model_id
minimax/MiniMax-M3*
pricing_scope
Literal source field names and values retained; provider, region, cache/batch/threshold conditions and per-unit meaning must be checked. max_tokens is not automatically a context limit.*
input cost per token
3.0e-7*
Scope and conditions
native_context
provider
minimax*
mode
chat*
full_native_model_id
minimax/MiniMax-M3*
pricing_scope
Literal source field names and values retained; provider, region, cache/batch/threshold conditions and per-unit meaning must be checked. max_tokens is not automatically a context limit.*
input cost per token above 512k tokens
6.0e-7*
Scope and conditions
native_context
provider
minimax*
mode
chat*
full_native_model_id
minimax/MiniMax-M3*
pricing_scope
Literal source field names and values retained; provider, region, cache/batch/threshold conditions and per-unit meaning must be checked. max_tokens is not automatically a context limit.*
litellm provider
minimax*
Scope and conditions
native_context
provider
minimax*
mode
chat*
full_native_model_id
minimax/MiniMax-M3*
pricing_scope
Literal source field names and values retained; provider, region, cache/batch/threshold conditions and per-unit meaning must be checked. max_tokens is not automatically a context limit.*
max input tokens
1,000,000*
Scope and conditions
native_context
provider
minimax*
mode
chat*
full_native_model_id
minimax/MiniMax-M3*
pricing_scope
Literal source field names and values retained; provider, region, cache/batch/threshold conditions and per-unit meaning must be checked. max_tokens is not automatically a context limit.*
max output tokens
128,000*
Scope and conditions
native_context
provider
minimax*
mode
chat*
full_native_model_id
minimax/MiniMax-M3*
pricing_scope
Literal source field names and values retained; provider, region, cache/batch/threshold conditions and per-unit meaning must be checked. max_tokens is not automatically a context limit.*
mode
chat*
Scope and conditions
native_context
provider
minimax*
mode
chat*
full_native_model_id
minimax/MiniMax-M3*
pricing_scope
Literal source field names and values retained; provider, region, cache/batch/threshold conditions and per-unit meaning must be checked. max_tokens is not automatically a context limit.*
output cost per token
1.2e-6*
Scope and conditions
native_context
provider
minimax*
mode
chat*
full_native_model_id
minimax/MiniMax-M3*
pricing_scope
Literal source field names and values retained; provider, region, cache/batch/threshold conditions and per-unit meaning must be checked. max_tokens is not automatically a context limit.*
output cost per token above 512k tokens
2.4e-6*
Scope and conditions
native_context
provider
minimax*
mode
chat*
full_native_model_id
minimax/MiniMax-M3*
pricing_scope
Literal source field names and values retained; provider, region, cache/batch/threshold conditions and per-unit meaning must be checked. max_tokens is not automatically a context limit.*
supports function calling
Yes*
Scope and conditions
native_context
provider
minimax*
mode
chat*
full_native_model_id
minimax/MiniMax-M3*
pricing_scope
Literal source field names and values retained; provider, region, cache/batch/threshold conditions and per-unit meaning must be checked. max_tokens is not automatically a context limit.*
supports prompt caching
Yes*
Scope and conditions
native_context
provider
minimax*
mode
chat*
full_native_model_id
minimax/MiniMax-M3*
pricing_scope
Literal source field names and values retained; provider, region, cache/batch/threshold conditions and per-unit meaning must be checked. max_tokens is not automatically a context limit.*
supports reasoning
Yes*
Scope and conditions
native_context
provider
minimax*
mode
chat*
full_native_model_id
minimax/MiniMax-M3*
pricing_scope
Literal source field names and values retained; provider, region, cache/batch/threshold conditions and per-unit meaning must be checked. max_tokens is not automatically a context limit.*
supports system messages
Yes*
Scope and conditions
native_context
provider
minimax*
mode
chat*
full_native_model_id
minimax/MiniMax-M3*
pricing_scope
Literal source field names and values retained; provider, region, cache/batch/threshold conditions and per-unit meaning must be checked. max_tokens is not automatically a context limit.*
supports tool choice
Yes*
Scope and conditions
native_context
provider
minimax*
mode
chat*
full_native_model_id
minimax/MiniMax-M3*
pricing_scope
Literal source field names and values retained; provider, region, cache/batch/threshold conditions and per-unit meaning must be checked. max_tokens is not automatically a context limit.*
supports vision
Yes*
Scope and conditions
native_context
provider
minimax*
mode
chat*
full_native_model_id
minimax/MiniMax-M3*
pricing_scope
Literal source field names and values retained; provider, region, cache/batch/threshold conditions and per-unit meaning must be checked. max_tokens is not automatically a context limit.*

* Provisional data