Best LLM API for text classification and ticket routing
What the benchmark found. Fireworks AI, Parasail, Telnyx, and Together AI tied for the task-success lead at 98.5%. The leading score represents approximately 197 exact successful results from 200 requests; operational failure rates were Fireworks AI 0.0%; Parasail 0.5%; Telnyx 0.0%; Together AI 0.0%. This result is based on 200 ticket-triage submissions per provider, not the pooled lookup and extraction rows. Measured across 10 LLM inference providers serving GLM 5.3 Flash, 200 requests each, Sep 3, 2026.
Providers measured. Baseten, DeepInfra, Fireworks AI, Modal, Nebius, Novita AI, Parasail, Telnyx, Together AI, and Z.AI, each serving GLM 5.3 Flash behind an OpenAI-compatible chat-completions endpoint.
What this page compares. This view isolates the benchmark's 200 ticket-triage questions instead of reusing the pooled three-task score. Each response must return both an exact category and an exact priority, making the task a measured proxy for short text classification and intent-routing decisions.
How to read the result. Use task success as the first gate when the classification changes a queue, policy, or next action. The percentage includes all submitted ticket cases; failure rate then separates provider delivery failures from HTTP-successful answers whose class or priority was wrong.
Workload and configuration. Every provider received the same 600 deliberately easy structured tasks: 200 meeting-note lookups, 200 ticket-triage classifications, and 200 contract-term extractions. GLM 5.3 Flash ran with streaming enabled, reasoning low, temperature 0, top-p 1, a 256-token output ceiling, one attempt, a 20-second timeout, and per-provider concurrency one.
Task-specific table. The table below uses only the published ticket-triage aggregate rather than the three-family pooled summary. Each provider therefore contributes 200 submitted requests to this view.
APIs used. The runner called each provider's OpenAI-compatible chat-completions surface with the model identifiers below; the links open the provider documentation used to verify each adapter.
- Fireworks AI:
accounts/fireworks/models/glm-5p3-flash· API docs ↗ - Parasail:
zai-org/GLM-5.3-Flash· API docs ↗ - Telnyx:
zai-org/GLM-5.3-Flash· API docs ↗ - Together AI:
zai-org/GLM-5.3-Flash· API docs ↗ - Baseten:
zai-org/GLM-5.3-Flash· API docs ↗ - Modal:
zai-org/GLM-5.3-Flash· API docs ↗ - DeepInfra:
zai-org/GLM-5.3-Flash· API docs ↗ - Z.AI:
glm-5.3-flash· API docs ↗ - Novita AI:
zai-org/glm-5.3-flash· API docs ↗ - Nebius:
zai-org/GLM-5.3-Flash· API docs ↗
Which LLM API classifies tickets most accurately?
This table contains only ticket-triage aggregates. Exact successes are shown out of 200, with failure rate and end-to-end p99 retained so accuracy can be balanced against operational reliability and blocking latency.
| Provider / API | Task success | Exact successes | Failure rate | End-to-end p99 |
|---|---|---|---|---|
| Fireworks AIaccounts/fireworks/models/glm-5p3-flash ↗ | 98.5% | 197 / 200 | 0.0% | 7.70 s |
| Parasailzai-org/GLM-5.3-Flash ↗ | 98.5% | 197 / 200 | 0.5% | 13.1 s |
| Telnyxzai-org/GLM-5.3-Flash ↗ | 98.5% | 197 / 200 | 0.0% | 3.78 s |
| Together AIzai-org/GLM-5.3-Flash ↗ | 98.5% | 197 / 200 | 0.0% | 6.83 s |
| Basetenzai-org/GLM-5.3-Flash ↗ | 98.0% | 196 / 200 | 0.0% | 1.20 s |
| Modalzai-org/GLM-5.3-Flash ↗ | 98.0% | 196 / 200 | 0.0% | 2.76 s |
| DeepInfrazai-org/GLM-5.3-Flash ↗ | 96.5% | 193 / 200 | 2.0% | 17.1 s |
| Z.AIglm-5.3-flash ↗ | 96.5% | 193 / 200 | 0.5% | 5.75 s |
| Novita AIzai-org/glm-5.3-flash ↗ | 95.0% | 190 / 200 | 3.5% | 4.11 s |
| Nebiuszai-org/GLM-5.3-Flash ↗ | 84.5% | 169 / 200 | 14.5% | 15.0 s |
Provider by provider, in the order above
- Fireworks AI: 98.5% task success (197 of 200 exact), 0.0% failures, end-to-end p99 7.70 seconds.
- Parasail: 98.5% task success (197 of 200 exact), 0.5% failures, end-to-end p99 13.12 seconds.
- Telnyx: 98.5% task success (197 of 200 exact), 0.0% failures, end-to-end p99 3.78 seconds.
- Together AI: 98.5% task success (197 of 200 exact), 0.0% failures, end-to-end p99 6.83 seconds.
- Baseten: 98.0% task success (196 of 200 exact), 0.0% failures, end-to-end p99 1.20 seconds.
- Modal: 98.0% task success (196 of 200 exact), 0.0% failures, end-to-end p99 2.76 seconds.
- DeepInfra: 96.5% task success (193 of 200 exact), 2.0% failures, end-to-end p99 17.07 seconds.
- Z.AI: 96.5% task success (193 of 200 exact), 0.5% failures, end-to-end p99 5.75 seconds.
- Novita AI: 95.0% task success (190 of 200 exact), 3.5% failures, end-to-end p99 4.11 seconds.
- Nebius: 84.5% task success (169 of 200 exact), 14.5% failures, end-to-end p99 14.99 seconds.
Which classification workflows are closest to the measured task?
- Support ticket triage. Route an inbound ticket only when both its issue category and urgency level are correct.
- Intent routing. Map short user text to a bounded intent enum before choosing a handler, tool, or workflow branch.
- Priority assignment. Return a structured severity label that downstream SLA and escalation rules can consume without parsing prose.
How is LLM inference provider latency and accuracy measured here?
- Same model. Every endpoint serves GLM 5.3 Flash; the provider layer is the only variable.
- Same request. One-turn, non-tool chat completion in JSON-object mode, streaming on, temperature 0, top-p 1, reasoning low, 256-token ceiling, 20-second timeout, concurrency one.
- Client clock, pinned tokenizer. Time to first token, end-to-end latency and tokens per second come from the client; visible answer tokens are counted with a pinned
tiktoken o200k_basetokenizer, never provider-reported usage. - Exact task success. An answer counts only when it is schema-valid and exactly correct under the task rubric. Timeouts, HTTP errors and malformed responses stay in the denominator.
- Serverless endpoints only. Dedicated capacity is outside the comparison. Runner and per-request records: openbenchmarks-labs/inference ↗.
Classification provider questions from the ticket task
Which provider achieved the highest exact success on the ticket-triage family?
Fireworks AI, Parasail, Telnyx, and Together AI tied for the task-success lead at 98.5%. The leading score represents approximately 197 exact successful results from 200 requests; operational failure rates were Fireworks AI 0.0%; Parasail 0.5%; Telnyx 0.0%; Together AI 0.0%. This result is based on 200 ticket-triage submissions per provider, not the pooled lookup and extraction rows.
What must a classification response return to count as correct?
A ticket-triage response counts as correct only when both the category and priority exactly match the reference enums. A correct category with the wrong priority receives no partial task-success credit.
How far can this ticket benchmark generalize to a private intent-routing taxonomy?
Intent routing is represented only by the benchmark's bounded ticket-triage task. It is useful evidence for short structured classification, but it does not test a production team's private intent taxonomy, class balance, or escalation policy.
Which LLM API is fastest for classification and ticket routing?
The end-to-end p99 column on this table is computed from the ticket-triage family alone. A synchronous classifier blocks the next branch until the complete category-and-priority object arrives, so p99 is the figure that protects a routing SLA; read it beside exact task success, because a fast wrong label does not finish the job.
Same run, other questions
- LLM inference provider benchmark: complete scorecard and methodology →
- Best LLM API for intent routing and ticket triage →
- Fastest LLM API for classification and ticket routing →
- Best LLM API for customer support automation →
- Best LLM inference provider for small tasks and short prompts →
- Fastest LLM inference provider for small tasks →
- Best LLM inference provider for enterprise AI programs →
Read the complete LLM inference provider benchmark
Metric definitions, request scheduling, task rubrics, failure handling, and the complete provider scorecard live on the LLM Inference Provider Benchmark →