BenchLeader
Multiverse ComputingReasoning modelAuto-detected

Quasar 438B (max, based on GLM-5.2)

Not yet ranked: too few independent results so far. Released 10 Aug 2026.

Blended price
$0.900/M
$0.600 in · $1.80 out
Output speed
177 tok/s
First answer
12 s
Context
1M
How it scores by categoryDashed line = average model (50). One step of 15 = one standard deviation.
  1. Knowledge63
  2. Long context65
  3. Composite63

Benchmark results

One column per reasoning effort. Rank is among every configuration of every model on that benchmark. Hover a score for the run it came from.

Reasoning

BenchmarkdefaultSource
GPQA Diamond (AA)not in index73.2%#262Artificial Analysis
Humanity's Last Exam (AA)not in index18.7%#170Artificial Analysis

Coding

BenchmarkdefaultSource
SciCode (AA)not in index48.1%#79Artificial Analysis

Knowledge

BenchmarkdefaultSource
AA-Omniscience-2.6#106Artificial Analysis

Long context

BenchmarkdefaultSource
AA-LCR76.3%#100Artificial Analysis

Composite

BenchmarkdefaultSource
AA Intelligence Index27.1#100Artificial Analysis

What a task costs

Estimates from list price, output speed and time to first answer for the best configuration. “With caching” assumes three-quarters of the input is served from the prompt cache. Reasoning tokens are not modelled.

WorkloadTokens in / outCostWith cachingTime
Chat reply400 / 300$0.000813.9 s
Summarise a 30-page report12,000 / 600$0.008315.6 s
Code edit6,000 / 1,500$0.006320.7 s
Agentic coding session60,000 / 4,000$0.04334.8 s
Structured extraction2,000 / 200$0.001613.3 s

See also

Data as of 9 Sept 2026. Compare with another model.