Skip to content

Ostra Pro

An advanced model for testing software. It writes and fixes tests, analyses failures, and reads screenshots and video.

Benchmarks

Ostra Pro against the coding models teams use for testing work today, on published price and on what a real testing request costs.

Cost to fix failing tests
  • Ostra Pro
  • Other coding models
Failing tests fixed against total cost
Model102030405060708090100
Ostra Pro$0.41$0.82$1.23$1.64$2.05$2.46$2.87$3.28$3.68$4.09
Claude Sonnet 5.5$1.20$2.40$3.60$4.80$6.00$7.20$8.40$9.60$10.80$12.00
Claude Opus 5.5$2.40$4.80$7.20$9.60$12.00$14.40$16.80$19.20$21.60$24.00
GPT-5.6 Terra$1.24$2.48$3.72$4.96$6.20$7.44$8.68$9.92$11.16$12.40
Grok 4.7$1.12$2.24$3.36$4.48$5.60$6.72$7.84$8.96$10.08$11.20

One request to fix a failing test: 50,000 input tokens and 2,000 output tokens. Each dot is the total cost of that many fixes, 10 to 100. Other models at their published API list prices, read 2026-11-19: Claude Sonnet 5.5, Claude Opus 5.5, GPT-5.6 Terra, Grok 4.7. No caching or batch discounts on either side.

Cost of input tokens
  • Ostra Pro
  • Other coding models
Input tokens against total cost
Model0M1M2M3M4M5M
Ostra Pro$0.00$0.64$1.28$1.92$2.56$3.20
Claude Sonnet 5.5$0.00$2.00$4.00$6.00$8.00$10.00
Claude Opus 5.5$0.00$4.00$8.00$12.00$16.00$20.00
GPT-5.6 Terra$0.00$2.00$4.00$6.00$8.00$10.00
Grok 4.7$0.00$2.00$4.00$6.00$8.00$10.00

Total cost of 1M to 5M input tokens at each model's list price per 1M, read 2026-11-19. Ostra Pro is $0.64.

Cost of output tokens
  • Ostra Pro
  • Other coding models
Output tokens against total cost
Model0M1M2M3M4M5M
Ostra Pro$0.00$4.47$8.94$13.41$17.88$22.35
Claude Sonnet 5.5$0.00$10.00$20.00$30.00$40.00$50.00
Claude Opus 5.5$0.00$20.00$40.00$60.00$80.00$100.00
GPT-5.6 Terra$0.00$12.00$24.00$36.00$48.00$60.00
Grok 4.7$0.00$6.00$12.00$18.00$24.00$30.00

Total cost of 1M to 5M output tokens at each model's list price per 1M, read 2026-11-19. Ostra Pro is $4.47.

Speed

0.3 s
Time to first token

Warm

157
Output tokens per second

One stream

10,128
Input tokens per second

Prompt processing

24–39 s
Cold start

To first token

Throughput under load
  • All streams
  • Per stream
Concurrent streams against output tokens per second
Model116
All streams1571,442
Per stream15790

Load test on one GPU with a warm container, 2026-11-17. Sixteen concurrent streams produce about nine times the output of one, at roughly 90 tokens per second each. One and sixteen streams were measured; the line joins the two.

What caching saves
$0.0000$0.050$0.10$0.15050K100K150K200KInput tokens per requestNo cache: $0.14 at 200K$0.14No cache50% cached: $0.079 at 200K$0.07950% cached90% cached: $0.033 at 200K$0.03390% cached

Ostra Pro only. Cost of a request with 2,000 output tokens, by prompt size and the share of the prompt served from cache.

Capabilities

What goes in, what it works from, what it does and what comes out.

Input

  • Text

    Logs, traces, history

  • Images

    Screenshots

  • Video

    Recordings

Context

  • Application memory

    Flaky flows, renamed selectors

  • Your suite

    Your framework and style

Tasks

  • Writes tests

    In your framework

  • Fixes failures

    Finds the cause, proposes a patch

  • Analyses and reasons

    Why a flow broke

  • Reads the screen

    Screenshots and video as evidence

Reasons or answers directlyCalls tools

Output

  • Streamed text

    As it is produced

  • New specs

    In your framework

  • Patches

    For you to approve

Built for the whole testing loop

Writing tests

New specs in the framework you already run, shaped by how your suite is written today.

Fixing failures

Reads the failing run, finds the cause and proposes the patch for you to approve.

Analysis and reasoning

Reads logs, traces and failure history, and works out why a flow broke before touching code.

Screenshots and video

Treats what was on screen as evidence, so a visual failure is something it can reason about.

Built for more than writing tests

Most of testing is not typing a spec. It is deciding whether a failure is real, finding what changed, and knowing which of a hundred tests is worth running. Ostra Pro is built for that work as well as for the code.

It reads TestRelic's application memory on every call, so it starts from what is true about your app: which flows flake, which selectors were renamed, and what your team has already decided.

Available everywhere you work

The same model answers in Studio, in your terminal, inside any MCP client and over REST.

Studio

Pick Ostra as the model and watch the agent work a failing run, step by step.

Open Studio
checkout.spec.ts failed on main · 3 of the last 10 runs
 
✓ Read the run — video, steps, network, console
✓ Memory — [data-test=pay-now] renamed in v2.4.1
● Patch the selector and re-run the spec
 
- await page.click('[data-test=pay-now]')
+ await page.click('[data-test=pay-submit]')
 
Waiting for your approval before anything touches disk.

CLI

The same agent in your terminal and, headless, in your pipeline.

Install the CLI
❯ why did checkout fail on staging last night?
 
⏺ memory.search done
⏺ runs.get_failures done
 
The selector [data-test=pay-now] was renamed to
[data-test=pay-submit] in v2.4.1. Two specs still use the old one.
 
model ostra-pro · shop-e2e

MCP

One server for Cursor, Claude Code, Copilot and Codex. Ostra shows up as tools.

Connect an agent
// .mcp.json — Cursor, Claude Code, Copilot, Codex
{
"mcpServers": {
"ostra": {
"type": "http",
"url": "https://ostra.testrelic.ai/mcp",
"headers": { "Authorization": "Bearer $OSTRA_API_KEY" }
}
}
}

API

REST with streaming. One bearer key for every model and every task.

Read the rates
$ curl https://ostra.testrelic.ai/api/v1/chat/completions \
-H "Authorization: Bearer $OSTRA_API_KEY" \
-d '{"model":"ostra-pro","stream":true,
"messages":[{"role":"user",
"content":"Fix the flaky checkout spec"}]}'
 
data: {"choices":[{"delta":{"content":"The flake comes from"}}]}
data: {"choices":[{"delta":{"content":" a renamed selector"}}]}

Try Ostra Pro now.

Frequently asked questions

Ostra Pro is TestRelic's advanced model for testing software. It writes and fixes tests, analyses failures, reasons about a flow, and reads screenshots and video. Teams use it where they would otherwise use Claude or Codex for testing.

In Studio, in the CLI, over MCP from Cursor, Claude Code, Copilot or Codex, and through the REST API at ostra.testrelic.ai/api/v1. All four use the same API key and the same prepaid balance.

Text, images and video go in, and text comes out, streamed as it is produced. It can reason, call tools, and answer directly.

Send a POST to https://ostra.testrelic.ai/api/v1/chat/completions with the model id ostra-pro and your API key as a bearer token. Responses stream. The same key works in Studio, the CLI and over MCP.

$0.64 per 1M input tokens, $0.06 per 1M cached input tokens and $4.47 per 1M output tokens, from a prepaid balance. Top-ups start at $5. You can pay as you go with no plan, and Ostra usage is also included in Growth, Pro and Enterprise.

No. You can pay as you go from a prepaid balance, with no plan. Growth, Pro and Enterprise also include Ostra usage, which is metered and resets automatically: Pro has 10x Growth's, and Enterprise is unlimited.

A request with no balance is refused until you add credit. You can add $5 to $999 per payment. To limit spend, each key can carry a spend cap, and the account has a monthly spend limit.

On price, a request to fix a failing test (50,000 input tokens, 2,000 output tokens) costs about $0.041 on Ostra Pro, against $0.11 to $0.24 on Claude Sonnet 5.5, Claude Opus 5.5, GPT-5.6 Terra and Grok 4.7 at their published list prices. We have not published head-to-head quality results yet.

In a load test on one GPU with a warm container (2026-11-17), the first token took about 0.3 seconds and one stream produced 157 output tokens per second. Sixteen concurrent streams produced 1,442 in total, about 90 each. Your numbers will vary with prompt size and concurrency.

Yes, when no capacity is warm. A request then waits roughly 24 to 39 seconds for the first token, and the API sends keep-alives while it waits. Call /api/v1/warm before a batch job to start capacity ahead of the first request.

Yes. It reads TestRelic's application memory on every call, so it starts from what is true about your app: which flows flake, which selectors were renamed, and what your team has already decided.

Use Ostra Decision when a pipeline or an agent needs a fast verdict: why a test failed, which tests a diff affects, or whether a selector will survive. It returns a verdict, not an explanation. When you need the reasoning or the fix, send the failure to Ostra Pro.