Skip to main content
AI Choice Engine

Model comparison

Claude Sonnet 5.5 vs GLM 5.3 Flash

Anthropic against Z.ai, compared on context, price, and attributed benchmark results.

Claude Sonnet 5.5: profile checked October 25, 2026GLM 5.3 Flash: profile checked October 25, 2026Individual provider fields may changeEvidence confidence: Medium — see receipts below

Anthropic

Claude Sonnet 5.5

Balanced

vs

Z.ai

GLM 5.3 Flash

Balanced · Open weights

AI model capability comparison
SpecificationClaude Sonnet 5.5GLM 5.3 Flash
ProviderAnthropicZ.ai
TierBalancedBalanced
Context window1MWinner: 1.05M
Max output128KNot verifiedUnverified
Input / 1M tokens$2Winner: $0.15
Output / 1M tokens$10Winner: $0.50
WeightsClosedOpen
ParametersNot verifiedUnverified320B total / 18B active (MoE)
Reasoning levelslow, medium, high, xhigh, maxlow, high, max
Modalitiestext, imagetext, image, video, file
LicenseNot verifiedUnverifiedMIT
API model idclaude-sonnet-5-5glm-5.3-flash
ReleasedSeptember 28, 2026August 26, 2026
Artificial Analysis Intelligence Index [max] · source date Claude Sonnet 5.5: 2026-10-0156Not verifiedUnverified
Artificial Analysis Intelligence Index · source date GLM 5.3 Flash: 2026-10-01Not verifiedUnverified42
Terminal-Bench 4.0 [max] · source date Claude Sonnet 5.5: 2026-10-0163.6Not verifiedUnverified
Terminal-Bench 4.0 · source date Claude Sonnet 5.5: 2026-10-0170.6Not verifiedUnverified
CursorBench 4.0 [max] · source date Claude Sonnet 5.5: 2026-10-0155.5Not verifiedUnverified
Humanity's Last Exam · source date Claude Sonnet 5.5: 2026-10-0164.5Not verifiedUnverified
Humanity's Last Exam · source date GLM 5.3 Flash: 2026-08Not verifiedUnverified55.3
FrontierCode 1.1 (Main) [xhigh] · source date Claude Sonnet 5.5: 2026-10-0152.1Not verifiedUnverified
OSWorld 2.1 · source date Claude Sonnet 5.5: 2026-10-0180.1Not verifiedUnverified
Terminal-Bench 2.1 · source date GLM 5.3 Flash: 2026-08Not verifiedUnverified84.3
DeepSWE 1.1 · source date GLM 5.3 Flash: 2026-08Not verifiedUnverified63.4

Dates beside benchmark scores refer to the linked source’s publication or capture date unless that source explicitly states when it measured the result. A shared URL alone does not establish comparable runs; unverified setups are marked. Prices are USD per million tokens at standard rates, excluding batch and caching discounts. Values we could not confirm from the provider are shown as “Not verified” rather than estimated.

Benchmark receipts

Where each score comes from, and how far it can be compared across models.

  • Source date2026-10-01
    Artificial Analysis (GLM-5.3-Flash)

    Composite index blending reasoning, knowledge, and coding evals into one 0–100 score.

    Comparable with caveatOnly same-version scores are comparable. v4.3.2 was rebaselined (the top score fell from 66 on v4.1.1 to ~58), so v4.1.1 figures must not be compared with v4.3.2 figures — check the cited source and its date.

  • Source date2026-10-01
    Anthropic (vendor-reported) (partial task set)
  • Source date2026-10-01
    Anthropic (vendor-reported)
  • Source date2026-10-01
    Anthropic (vendor-reported) (with tools)

    Frontier-knowledge ceiling: extremely hard multi-domain questions written to be near-impossible without deep expertise.

    Comparable with caveatSubset/tool configurations (text-only vs with-tools) differ between vendors and materially change scores; must match configuration to compare.

  • Source date2026-10-01
    Cursor (vendor-run coding agent)
  • Source date2026-10-01
    Artificial Analysis (independent, mini-swe-agent)

    Agentic terminal work: long multi-step tasks executed in a sandboxed shell environment.

    Comparable with caveatDo not directly compare scores across Terminal-Bench versions: 4.0 revises and prunes the 3.0 task set and changes environment resources, requiring fresh trials. Within 4.0, harness, time limit, tools, sampling and resource settings must match; maintainer-agent and Artificial Analysis runs are not interchangeable.

  • Source date2026-10-01
    Anthropic (vendor-reported)

    Agentic terminal work: long multi-step tasks executed in a sandboxed shell environment.

    Comparable with caveatDo not directly compare scores across Terminal-Bench versions: 4.0 revises and prunes the 3.0 task set and changes environment resources, requiring fresh trials. Within 4.0, harness, time limit, tools, sampling and resource settings must match; maintainer-agent and Artificial Analysis runs are not interchangeable.

  • Source date2026-10-01
    Artificial Analysis (Claude Sonnet 5.5 (max with fallback))

    Composite index blending reasoning, knowledge, and coding evals into one 0–100 score.

    Comparable with caveatOnly same-version scores are comparable. v4.3.2 was rebaselined (the top score fell from 66 on v4.1.1 to ~58), so v4.1.1 figures must not be compared with v4.3.2 figures — check the cited source and its date.

  • Source date2026-08
    GLM-5.3-Flash HF card (vendor, tools/full set; 300K context, 163,840 max generation, GPT-5.6-Luna medium judge)

    Frontier-knowledge ceiling: extremely hard multi-domain questions written to be near-impossible without deep expertise.

    Comparable with caveatSubset/tool configurations (text-only vs with-tools) differ between vendors and materially change scores; must match configuration to compare.

  • Source date2026-08
    Z.ai GLM-5.3-Flash blog (vendor, mini-swe-agent, 400K context)

    Agentic software engineering: pass@1 on real repo tasks with reported cost per completed task.

    Directly comparable

  • Source date2026-08
    GLM-5.3-Flash HF card (vendor; Claude Code 2.1.207, 65,536 max generation, 6h timeout)

    Agentic terminal work: multi-step tasks executed in a sandboxed shell environment.

    Comparable with caveatNot comparable with Terminal-Bench 3.0 or 4.0 (different task sets) or v1; harness configuration (container, time limits) also shifts results.

Pricing tiers

Claude Sonnet 5.5: Standard USD/MTok input/output $2/$10 with no long-context surcharge; cache reads $0.10, 5-minute writes $2.50, 1-hour writes $4. Batch $1/$5. US-only inference adds 10% to input/output and cache reads/writes on Anthropic-operated routes where supported; Foundry uses US Data Zone Standard. Bedrock and Google-managed Google Cloud have independent regional prices. Batch 300K output requires output-300k-2026-03-24; standard output stays 128K. Knowledge cutoff June 2026; retirement not sooner than September 28, 2027. Server-side tools can add usage fees; tool schemas/results and thinking consume billable tokens.

GLM 5.3 Flash: Z.ai PAYG USD/MTok input/output $0.15/$0.50, cached input $0.03; cache storage is limited-time free with no stated end date. Web Search is separately $0.01/use. Hosted guide advertises 1M context and 128K output, while the native checkpoint config fixes 1,048,576 positions. An exact Flash-specific integer output ceiling remains unverified. FlashX has separate $0.37/$1.25 pricing and $0.075 cache reads.

BalancedRecord checked October 25, 2026

Claude Sonnet 5.5

Anthropic’s newer Sonnet model for everyday coding and knowledge work at the same $2/$10 token rates as Sonnet 5.

Best for

  • Everyday coding
  • Document and slide creation
  • Fast iteration

Watch out

Vendor scores keep their original evaluation setup and dates. between_tools turns off up-front thinking only at low/medium/high effort; xhigh/max requires adaptive thinking. Changing effort per message while between_tools is active returns an error. Forced tool use, temperature/top_p/top_k, older computer_20251124 on Claude API/Google-managed Google Cloud, and some advisor pairings are rejected. Thinking blocks are model/conversation-bound; progress text needs a display setting that returns it. Haiku 5.5 has now released separately; its capabilities and scores are not inherited.

BalancedOpen weightsRecord checked October 25, 2026

GLM 5.3 Flash

Z.ai's multimodal MIT-licensed 320B/18B MoE checkpoint and hosted Flash endpoint, with always-enabled reasoning and publisher coding/agent evaluations.

Best for

  • Cost-efficient long-context
  • Multimodal input
  • Coding agents

Watch out

The 320B weights require substantial hardware; memory depends on quantization, runtime and context. Effort low/high/max defaults to max. Flash is on Coding Plan with three times GLM-5.3 quota; FlashX is not on the plan. Half-point off-peak/weekend use applies to plan quota, not per-token API prices. Publisher evaluation generation budgets are not hosted API output caps. The shared guide recommends temperature=1/top_p=0.95, preserving thinking with clear_thinking=false, and tool_stream=true for streamed tool calls. File/image/video understanding is input capability, not a native media-generation output contract.

When the cheaper one wins

GLM 5.3 Flash is cheaper on output at $0.50 per million tokens against $10 for Claude Sonnet 5.5 — about 20×. Use the cheaper tier for classification, extraction, summarisation, and any task where the expensive model’s extra score does not change the accepted output. The expensive one only pays if your hardest task actually fails on the cheap tier. This compares standard short-context token rates, not total cost per accepted task. Include input, cache, context tiers, schedules, retries and review in your pilot.

Run the model picker

Evidence confidence: Medium

How strong and complete the evidence behind this comparison is — not a prediction of which model is better.

  • Pricing verified on both sides — Input and output rates are verified for both models.
  • 3/5 core specs verified on both sides — Not published for at least one side: max output, parameter count.
  • No directly aligned benchmark setup — No benchmark has matching cited sources, source dates, units, effort and recorded run details on both models.
  • Verified within the last 90 days — Older of the two catalog checks was 1 day ago.
  • Both models carry source citations — Each side has at least two catalog sources on record.

Source receipts

Each catalog figure was checked against the provider or an independent second source on the date shown.

Related comparisons

Diving deeper on one model? Claude Sonnet 5.5 · GLM 5.3 Flash

Common questions

Claude Sonnet 5.5 vs GLM 5.3 Flash

Answered from the verified figures on this page rather than general guidance.

Is Claude Sonnet 5.5 or GLM 5.3 Flash cheaper for input?
GLM 5.3 Flash is cheaper at $0.15 per million input tokens, against $2 for Claude Sonnet 5.5 — roughly 13× the price. Estimate both input and output token volume for your workload. These are base rates; Claude Sonnet 5.5: Standard USD/MTok input/output $2/$10 with no long-context surcharge; cache reads $0.10, 5-minute writes $2.50, 1-hour writes $4. Batch $1/$5. US-only inference adds 10% to input/output and cache reads/writes on Anthropic-operated routes where supported; Foundry uses US Data Zone Standard. Bedrock and Google-managed Google Cloud have independent regional prices. Batch 300K output requires output-300k-2026-03-24; standard output stays 128K. Knowledge cutoff June 2026; retirement not sooner than September 28, 2027. Server-side tools can add usage fees; tool schemas/results and thinking consume billable tokens. GLM 5.3 Flash: Z.ai PAYG USD/MTok input/output $0.15/$0.50, cached input $0.03; cache storage is limited-time free with no stated end date. Web Search is separately $0.01/use. Hosted guide advertises 1M context and 128K output, while the native checkpoint config fixes 1,048,576 positions. An exact Flash-specific integer output ceiling remains unverified. FlashX has separate $0.37/$1.25 pricing and $0.075 cache reads.
Is Claude Sonnet 5.5 or GLM 5.3 Flash cheaper for output?
GLM 5.3 Flash is cheaper at $0.50 per million output tokens, against $10 for Claude Sonnet 5.5 — roughly 20× the price. Estimate both input and output token volume for your workload. These are base rates; Claude Sonnet 5.5: Standard USD/MTok input/output $2/$10 with no long-context surcharge; cache reads $0.10, 5-minute writes $2.50, 1-hour writes $4. Batch $1/$5. US-only inference adds 10% to input/output and cache reads/writes on Anthropic-operated routes where supported; Foundry uses US Data Zone Standard. Bedrock and Google-managed Google Cloud have independent regional prices. Batch 300K output requires output-300k-2026-03-24; standard output stays 128K. Knowledge cutoff June 2026; retirement not sooner than September 28, 2027. Server-side tools can add usage fees; tool schemas/results and thinking consume billable tokens. GLM 5.3 Flash: Z.ai PAYG USD/MTok input/output $0.15/$0.50, cached input $0.03; cache storage is limited-time free with no stated end date. Web Search is separately $0.01/use. Hosted guide advertises 1M context and 128K output, while the native checkpoint config fixes 1,048,576 positions. An exact Flash-specific integer output ceiling remains unverified. FlashX has separate $0.37/$1.25 pricing and $0.075 cache reads.
Which has the larger context window, Claude Sonnet 5.5 or GLM 5.3 Flash?
GLM 5.3 Flash accepts 1.05M tokens against 1M for Claude Sonnet 5.5. This only matters if you routinely send very long documents or large codebases.
Do Claude Sonnet 5.5 and GLM 5.3 Flash support the same reasoning levels?
Claude Sonnet 5.5 exposes low, medium, high, xhigh, max, while GLM 5.3 Flash exposes low, high, max.
Should I use Claude Sonnet 5.5 or GLM 5.3 Flash?
Both are classified in our balanced tier; this editorial grouping does not imply equal capability. Claude Sonnet 5.5 is a candidate for everyday coding; GLM 5.3 Flash is a candidate for cost-efficient long-context. Compare accuracy, tool reliability, latency and cost on representative tasks.
Can I self-host Claude Sonnet 5.5 or GLM 5.3 Flash?
GLM 5.3 Flash publishes open weights, but self-hosting and third-party provider access depend on the licence, hardware, serving support, and availability. Claude Sonnet 5.5 is a closed model whose supported access paths are controlled by its provider.

Next step

Choosing between them

The verified figures that separate this pair, computed from the catalog rather than restated boilerplate.

  • Input price: GLM 5.3 Flash costs $0.15 per 1M tokens versus $2 for Claude Sonnet 5.5 — a 13.3x difference at the headline tier.
  • Context: GLM 5.3 Flash takes 1.05M against 1M for Claude Sonnet 5.5 — only decisive if your prompts approach the smaller window.
  • Deployment: GLM 5.3 Flash publishes weights you can self-host; the other is API-only.
Run the model pickerEstimate stack costAll model comparisons

Choosing a harness rather than a model — Cursor, Copilot, Claude Code, Windsurf, or Lovable? Compare agentic harnesses · Latest releases.