Z.ai, released Jul 28, 2025

GLM-4.5price, context, benchmarks and release details

Provisional: not enough results to rank yet Open weights 47% confidence 47 percent, Low confidence, 3 of 8 expected sources in
51.0
SI Score
Not ranked yet
Input, per 1M tokens
$0.60Z.ai API pricingOfficial Z.ai global Standard uncached token rate; coding-plan subscriptions, cache and Batch excluded Retrieved Oct 9, 2026 · factual citation
Open source ↗
Output, per 1M tokens
$2.20Z.ai API pricingOfficial Z.ai global Standard uncached token rate; coding-plan subscriptions, cache and Batch excluded Retrieved Oct 9, 2026 · factual citation
Open source ↗
Context window
131Kmodels.devPublished source fact Retrieved Oct 9, 2026 · MIT
Open source ↗
Max output
98.3Kmodels.devPublished source fact Retrieved Oct 9, 2026 · MIT
Open source ↗
Released
Jul 28, 2025models.devPublished source fact Retrieved Oct 9, 2026 · MIT
Open source ↗

How the score breaks down

Coding (weight 40 percent) 43.9
Math (weight 15 percent) —
Preference (weight 15 percent) 76.0
Reasoning (weight 30 percent) —

Weights: reasoning 30%, math 15%, coding 40%, preference 15%. Results use fixed 0–100 scales before averaging, and thin evidence is pulled toward 50. Method si-v3-retained-evidence-2, computed Oct 9, 2026, 06:15 UTC.

Around it on the leaderboard

  1. 1 Claude Fable 5.1 80.2
  2. 2 Claude Opus 5.5 78.4
  3. 3 GPT-6 Astra 77.5
  4. 4 Claude Fable 5 76.8
  5. 5 Claude Opus 5 74.9

Full leaderboard

Benchmark results

3 benchmarks, 4 results

Each row shows the best published result. Where a model was tested at several settings, such as reasoning effort, open the row to see each one. Hover or tap a value for its source.

SWE-bench Verifiedcoding 64.2%SWE-bench VerifiedSWE-bench published model plus agent result; harness retained, not a base model evaluation [variant] UndisclosedPublished Jul 28, 2025 Retrieved Oct 9, 2026 · factual citation
Open source ↗ 64.2 2 settings
  • Setting 1 54.2%SWE-bench VerifiedSWE-bench published model plus agent result; harness retained, not a base model evaluation [variant] mini-SWE-agent; 1.9.1Published Aug 22, 2025 Retrieved Oct 9, 2026 · factual citation
    Open source ↗
  • Setting 2 64.2%SWE-bench VerifiedSWE-bench published model plus agent result; harness retained, not a base model evaluation [variant] UndisclosedPublished Jul 28, 2025 Retrieved Oct 9, 2026 · factual citation
    Open source ↗
About SWE-bench Verified
Terminal-Benchcoding 22%Official model cards via models.devLab-reported; metric success rate; transcribed by MIT models.dev catalog; not independently evaluated [variant] Published Mar 11, 2026 Retrieved Oct 9, 2026 · factual citation; MIT transcription
Open source ↗ 22.0
LMArena Textpreference 1430.0 eloLMArena / ArenaPublished source fact [variant] text / overallPublished Oct 8, 2026 Retrieved Oct 9, 2026 · CC-BY-4.0
Open source ↗ 76.0

Normalization uses a fixed 0–100 scale for each unit, independent of other models. Compare evaluation conditions before reading a small gap as decisive. “Lab-reported” marks the provider's own published figure.

Details and sources

Open weights
Yesmodels.devPublished source fact Retrieved Oct 9, 2026 · MIT
Open source ↗
License
not yet reported
Input modalities
textmodels.devPublished source fact Retrieved Oct 9, 2026 · MIT
Open source ↗
First seen by SuperIndex
Oct 8, 2026
Coverage
38% of expected source weight

Reported (3)

  • Official model cards via models.devOct 8, 2026
  • LMArena / ArenaOct 8, 2026
  • SWE-bench VerifiedOct 8, 2026

Awaiting (5)

  • ARC Prize11% of weight
  • Epoch AI Benchmarking23% of weight
  • Humanity’s Last Exam11% of weight
  • LiveBench11% of weight
  • Terminal-Bench6% of weight

Confidence rises as pending sources publish. Some sources never cover some models, so confidence reaches 100% at 80% of expected weight.

What changed

All releases · RSS feed

Alerts on this device

What to be alerted about
RSS feed