LLM Reference

GLM-5 vs GPT-5.2 Codex

GLM-5 and GPT-5.2 Codex solve the same coding-agent problem from opposite deployment models. GLM-5 is Zhipu AI's open-source (MIT) 200K-context foundation model with a tracked SWE-bench Verified row and multiple provider routes. GPT-5.2 Codex is OpenAI's proprietary coding model for long-horizon agentic engineering, with a 400K-token API context window and official OpenAI pricing. Use this comparison to decide whether openness and routability matter more than OpenAI's managed Codex model path.

Pick GLM-5 when you need open-source (MIT), downloadable weights, self-hosting options, or provider portability, especially for teams that want to inspect or run the model outside a closed API. Pick GPT-5.2 Codex when you want OpenAI's API-hosted coding model with a larger 400K context window, image input, 128K max output, and official $1.75/M input and $14/M output pricing. Do not mark GPT-5.2 Codex unavailable just because older Codex sign-in model choices were deprecated; the API model page remains the source of truth for API availability.

Decision scorecard

Local evidence first
SignalGLM-5GPT-5.2 Codex
Product typeStandalone API modelCoding-specialized model
Best forreasoning-heavy apps, tool-calling agents, and provider-routed productioncustom coding agents, code generation, and tool loops
Decision fitCoding, RAG, and AgentsCoding, RAG, and Agents
Context window200k400k
Cheapest output$2.08/1M tokens$14/1M tokens
Provider routes7 tracked2 tracked
Shared benchmarksSWE-bench Verified leader2 shared

Decision tradeoffs

Choose GLM-5 when...
  • GLM-5 holds a shared-benchmark lead on SWE-bench Verified, ahead by 5 points.
  • GLM-5 has the lower cheapest tracked output price at $2.08/1M tokens.
  • GLM-5 has broader tracked provider coverage for fallback and procurement flexibility.
  • Local decision data tags GLM-5 for Coding, RAG, and Agents.
Choose GPT-5.2 Codex when...
  • GPT-5.2 Codex holds a shared-benchmark lead on SWE-bench Pro, ahead by 1.3 points.
  • GPT-5.2 Codex has the larger context window for long prompts, retrieval packs, or transcript analysis.
  • GPT-5.2 Codex uniquely exposes Vision, Multimodal, and Code execution in local model data.
  • Local decision data tags GPT-5.2 Codex for Coding, RAG, and Agents.

Monthly cost at traffic

Estimate token spend from the cheapest tracked input and output route or tier on this page.

Lower estimate GLM-5

GLM-5

$1,000

Cheapest tracked route/tier: OpenRouter

GPT-5.2 Codex

$4,900

Cheapest tracked route/tier: Vercel AI Gateway

Estimated monthly gap: $3,900. Batch, cache, alternate speed tiers, and negotiated pricing are excluded from this local estimate.

Switch friction

GLM-5 -> GPT-5.2 Codex
  • Provider overlap exists on Vercel AI Gateway; start route-level A/B tests there.
  • GPT-5.2 Codex is $11.92/1M tokens higher on cheapest tracked output pricing, so quality gains need to justify the spend.
  • GPT-5.2 Codex adds Vision, Multimodal, and Code execution in local capability data.
GPT-5.2 Codex -> GLM-5
  • Provider overlap exists on Vercel AI Gateway; start route-level A/B tests there.
  • GLM-5 is $11.92/1M tokens lower on cheapest tracked output pricing before cache, batch, or negotiated discounts.
  • Check replacement coverage for Vision, Multimodal, and Code execution before moving production traffic.

Specs

Specification
Released2026-02-112025-12-18
Context window200k400k
Parameters744B total, 40B active
ArchitectureMixture of ExpertsDecoder Only
LicenseMITOSI-approvedProprietary
OpennessOpen sourceProprietary
WeightsAvailableNot released
CodeUnknownUnknown
Commercial useCommercial use: permittedCommercial use: conditional
Knowledge cutoff2025-112025-08

Pricing and availability

Pricing attributeGLM-5GPT-5.2 Codex
Input price$0.60/1M tokens$1.75/1M tokens
Output price$2.08/1M tokens$14/1M tokens
Providers

Capabilities

CapabilityGLM-5GPT-5.2 Codex
VisionNoYes
MultimodalNoYes
ReasoningYesYes
Function callingYesYes
Tool useYesYes
Structured outputsYesYes
Code executionNoYes
IDE integrationNoNo
Computer useNoNo
Parallel agentsNoNo

Benchmarks

BenchmarkGLM-5GPT-5.2 Codex
SWE-bench Verified77.872.8
SWE-bench Pro55.156.4

Deep dive

The first decision is deployment control. GLM-5 is released open-source (MIT) with weights on Hugging Face and a Z.ai model page that describes it as a foundation model for agentic engineering. GPT-5.2 Codex is closed but has a current OpenAI API model page with pricing and endpoints.

Context now favors GPT-5.2 Codex on the checked primary sources: OpenAI lists 400,000 context and 128,000 max output tokens, while Z.ai lists GLM-5 at 200K context and 128K maximum output.

Pricing is provider-dependent for GLM-5 and direct for GPT-5.2 Codex. The safest snippet should cite the OpenAI API price for GPT-5.2 Codex and avoid a generic GLM-5 direct-token price unless the specific provider row is shown.

As an MIT-licensed open-source model, GLM-5 supports self-hosting via standard inference frameworks. Benchmark evidence is asymmetric: GLM-5 has a seeded SWE-bench Verified score from public leaderboard data. The checked OpenAI primary sources do not publish numeric GPT-5.2 Codex SWE-Bench Pro or Terminal-Bench 2.0 scores, so the page should say the numeric row is unavailable instead of implying a head-to-head score.

FAQ

Is GPT-5.2 Codex still available through the API?

Yes. OpenAI's API model page lists GPT-5.2-Codex with context, token pricing, modalities, and API endpoints. Codex docs separately note deprecations for some ChatGPT sign-in model choices, so treat API availability and Codex UI model selection as separate statuses.

Which has the larger context window, GLM-5 or GPT-5.2 Codex?

GPT-5.2 Codex has the larger checked API window at 400K tokens. GLM-5 is listed by Z.ai at 200K context. Both pages list a 128K maximum output/token cap.

What is GLM-5?

GLM-5 is Zhipu AI's open-source flagship model published under the OSI-approved MIT license, with weights downloadable on Hugging Face (https://huggingface.co/zai-org/GLM-5). It has a 200K-token context window and a tracked SWE-bench Verified row. GPT-5.2 Codex is proprietary and should be chosen when OpenAI API hosting, image input, and managed coding-agent behavior are more important than model portability.

Can I self-host GLM-5?

Yes. GLM-5 is open-source (MIT) and downloadable on Hugging Face (https://huggingface.co/zai-org/GLM-5). It supports self-hosting via standard inference frameworks. GPT-5.2 Codex is proprietary and not available for self-hosting.

Can I compare GLM-5 and GPT-5.2 Codex by SWE-bench score?

Only partially. GLM-5 has a tracked SWE-bench Verified row in the seed. The checked OpenAI primary sources do not provide a numeric GPT-5.2 Codex SWE-bench score, so the page should show the missing numeric row rather than inventing one.

Continue comparing

Last reviewed: 2026-06-30. Data sourced from public model cards and provider documentation.