Z.ai rolls GLM-5.3 to $18 Coding Plan users with major agent benchmark jumps
According to Towards AI reporting, Z.ai released GLM-5.3 to current $18-per-month Coding Plan subscribers, keeping the GLM-5.2 base while adding post-training improvements tuned for autonomous coding agents instead of general chat. Access arrives through automatic plan routing with a one-million-token context window and 128K output ceiling that Z.ai positions as subscription-bound agent infrastructure rather than a broad public API. For teams running multi-file agent pipelines, wider context may let one session cover an entire codebase without heavy chunking. Z.ai reports Terminal-Bench 3.0 climbed from 4.6 to 28.3 and DeepSWE from 46.2 to 66.9 on its internal benchmarks, but those gains are not independently verified in available reporting and wider API access remains undocumented.
Z.ai rolls GLM-5.3 to $18 Coding Plan users with major agent benchmark jumps
Z.ai made GLM-5.3 available to existing $18-per-month Coding Plan subscribers while keeping the GLM-5.2 base model and adding post-training gains. Z.ai reports Terminal-Bench 3.0 rose from 4.6 to 28.3 and DeepSWE from 46.2 to 66.9 on its internal benchmarks.
Key takeaway
GLM-5.3 signals a pivot toward subscription-gated, high-context agent infrastructure, prioritizing specialized coding workflows over general public API availability.
What happened
Z.ai made GLM-5.3 available to existing $18-per-month Coding Plan subscribers while keeping the GLM-5.2 base model and adding post-training gains, according to Towards AI reporting on the rollout.
The company positions GLM-5.3 with a 1M-token context window and 128K output limit for coding agents, available exclusively through the Coding Plan with automatic routing. Z.ai reports Terminal-Bench 3.0 rose from 4.6 to 28.3 and DeepSWE from 46.2 to 66.9 on its internal benchmarks.
Evidence
GLM-5.3 is available to existing $18-per-month Coding Plan subscribers without a plan migration.
Towards AI · attributed
Z.ai has made a potentially expensive coding-agent upgrade available to its existing $18-per-month plan users without asking them to move
GLM-5.3 retains the GLM-5.2 base with post-training refinements and a 1M-token context window.
Towards AI · attributed
Z.ai has introduced GLM-5.3, which retains the GLM-5.2 base but introduces a 1M-token context window and 128K output limit specifically for coding agents.
Z.ai reports large internal benchmark jumps for Terminal-Bench 3.0 and DeepSWE.
Towards AI · attributed
Z.ai reports Terminal-Bench 3.0 rose from 4.6 to 28.3 and DeepSWE from 46.2 to 66.9 on its internal benchmarks.
Z.ai is positioning GLM-5.3 as specialized agent infrastructure rather than a general-purpose chat model.
Towards AI · attributed
Z.ai is leveraging a 1M-token context window and post-training refinements to position GLM-5.3 as a specialized agent infrastructure layer rather than a general-purpose chat model.
Why it matters
For builders, the 1M-token context window enables processing entire large-scale codebases in a single pass, changing how context is managed in autonomous coding agents.
Limits and uncertainties
Terminal-Bench 3.0 and DeepSWE gains cited are Z.ai internal benchmarks and are not independently verified in the available reporting.
GLM-5.3 is described as available exclusively through the $18/month Coding Plan; broader public API availability is not documented in the packet.
Practical implications
Existing $18 Coding Plan subscribers receive GLM-5.3 via automatic routing without migrating plans.
Builders may need to evaluate Coding Plan access if they want GLM-5.3 rather than waiting for general API availability.
What to watch
Whether Z.ai opens GLM-5.3 beyond the Coding Plan to public API or other tiers.
Independent replication of Terminal-Bench 3.0 and DeepSWE scores outside Z.ai's internal benchmarks.
