CODING · MODELS
GLM-5.3-Flash for coding agents
A fast, India-resident model with a 1 million-token context, verified tool-calling, and the prefix cache that makes agent loops cheap. Here's where it sits next to the Claude/Codex tiers you know, the latency we actually measured, and how to point your coding agent at it in one line.
What it is
glm-5.3-flash is served on Krutrim Cloud and reachable on the BharatRouter platform key — no BYOK. It's MIT-licensed, runs fp8 in an India datacenter, and carries a 1,048,576-token context window (131,072 max output). We confirmed it emits proper OpenAI tool_calls live, so agentic harnesses (brcode, Codex, Claude Code, Hermes) drive it the same way they drive a frontier model.
In Claude/Codex parlance
Capability tiers are approximate and task-dependent — benchmarks aren't the same as how a model feels in your repo — but a useful frame:
| If you're used to… | Reach for… | Why |
|---|---|---|
| Claude Haiku (fast, cheap, high-volume) | glm-5.3-flash | Same speed/cost role — but a 1M window (vs ~200K) and real reasoning + tools. Great for whole-repo prompts and long files. |
| Claude Sonnet / Codex default (day-to-day agentic coding) | gpt-oss-120b | The heavier agentic-coding workhorse; the model Codex CLI ships open-weight support for. Pick this for long tool-loop tasks. |
| Claude Opus (frontier) | — | No exact platform-served match yet; the honest gap. GLM-5 full or a larger model on dedicated hardware is the closest. |
Short version: Haiku-class speed and cost, Sonnet-ish context and reasoning.For the heaviest step-by-step agentic coding, gpt-oss-120b still leads; for fast iteration and large context, glm-5.3-flash is the better pick.
The latency we measured
Measured from Bengaluru against api.bharatrouter.com — streamed, warm:
- Time to first token: ~0.62s (streaming).
- ~130 tokens/sec single-stream decode.
Under load, per-stream decode holds up well (Krutrim Cloud serving figures):
| Concurrent streams | 4 | 8 | 16 | 32 | 128 |
|---|---|---|---|---|---|
| tokens/sec each | 120 | 60 | 40 | 25 | 10 |
For interactive coding, 40 tok/s (≈16 people generating at once) is faster than you read; 25 tok/s (≈32 at once) is still comfortable. Reproduce the first-token number yourself:
{reproCurl}The cache you'll feel
Agentic coding resends the same system prompt + repo context on every tool-loop turn. A prefix cache reuses the already-computed attention state for that shared prefix instead of recomputing it — so the second, third and hundredth turn of a session start faster and cost less than the first. This is why a "flash" model comfortably carries long agent sessions: the expensive part (your big context) is paid once.
It's also why the tier table above understates glm-5.3-flash for agent work — the 1M window plus prefix reuse means you can keep a large codebase in context across a whole session without paying for it every turn.
How to use it
One br- key; set the model to glm-5.3-flash. Pick your harness.
Any OpenAI-compatible client (curl / Python / Hermes)
{curlEx}{pyEx}Hermes, or anything that takes a base URL + key + model, is the same three settings:
{hermesEx}brcode (Codex-/Claude-Code-style agent)
{brcodeEx}AGENTS.md travels with your repo unchanged. Full guide:the brcode reference · migrating fromCodex CLI orClaude Code.
Keep the real Codex CLI
BharatRouter speaks the Responses API, so the codex binary can route through it:
{codexEx}Metering & budgets
Every call is metered and audited on your key. Admins set ₹/month or token budgetsper org, team, member or key — and the caps enforce (a request over budget gets a clean 429), so usage stays governed. Set them on the console's Spend & Budgetspanel.
"Claude", "Codex" and "Sonnet/Haiku/Opus" are products of Anthropic and OpenAI; tier comparisons are approximate positioning, not head-to-head benchmarks. Latency numbers are measured from Bengaluru and will vary with your location and load.