GPT-6.1 Terra vs GPT-5.6 Terra: Benchmark Results across 72 Browser Tasks
GPT‑6.1 Terra is the better browser-agent model of the two and the cheaper one. GPT‑6.1 Terra passed 43 of 72 tasks (59.7%), against 27 of 72 (37.5%) for GPT‑5.6 Terra. It also cost $0.286 per task, against $0.742. It was also faster where the two can be compared: on the 25 tasks both models passed, it took 200.9 seconds per task on average, against 233.0 seconds for GPT‑5.6 Terra, and finished first on 17 of the 25.
The biggest GPT‑6.1 Terra vs GPT‑5.6 Terra difference is in multi-site data collection, where it passed 11 of 32 tasks against 2 of 32. Across all 72 tasks, it passed 18 that GPT‑5.6 Terra failed and failed two that GPT‑5.6 Terra passed. Most of those 18 came down to the final table, where GPT‑5.6 Terra had filled in a wrong value or written "not shown" for a value the site does show.
GPT‑6.1 Terra is cheaper to run, not more expensive. It cost less per task in every type of work we tested, by 59% to 71%. The saving comes from OpenAI's lower prices rather than from lighter use, since GPT‑6.1 Terra read 31% more input tokens per task. Use it for multi-site data collection, where it was more accurate and cheaper, and for web apps automation, where it was also faster. Outside web apps automation the speed picture is mixed: GPT‑5.6 Terra was faster on 6 of the 11 shared tasks and had the lower average.
Success rate
GPT‑6.1 Terra passed 59.7% of tasks (43 of 72) and GPT‑5.6 Terra 37.5% (27 of 72). Counted by individual scored items, it met 359 of 489 (73.4%) against 299 (61.1%). The gap is widest in multi-site data collection. In the other four types of work, the two models were one to three tasks apart.
Tasks passed, by type of work
What GPT‑6.1 Terra fixed was the final table. Among the 18 tasks that only GPT‑6.1 Terra passed, GPT‑5.6 Terra entered an incorrect table value in 14 and wrote "not shown" for a value the site actually displayed in 7. Its wrong values were usually the right kind of answer for the wrong thing, such as a later talk named as RubyKaigi's opening keynote. GPT‑6.1 Terra got those right, and it filled in values GPT‑5.6 Terra had given up on, such as Norges Bank's zero dissents.
Our earthquake task shows the difference in one row. It asks nine seismology agencies for their readings of the 2025 Kamchatka earthquake, including each one's distance from a named place. GPT‑5.6 Terra read the USGS event page once and wrote "not shown" for that distance. GPT‑6.1 Terra opened the page's Regional Information tab, found Petropavlovsk-Kamchatsky listed 125.4 km WNW of the epicenter, and wrote the "125.4 km ESE of Petropavlovsk-Kamchatsky" the task expects.
GPT‑6.1 Terra opens Regional Information to find the distance
ego-browser nodejs <<'EOF'const t=await taskSpace('eff4-high-earthquake-agencies-01-…');const p=t.page('p1');await p.click('@251');await p.waitForTimeout(2000);console.log((await p.snapshot({scope:'full_page'})).slice(0,16000));EOF@251 is the Regional Information link in the step 5 page snapshot.…heading text "Nearby Places"ordered_list list_item text "Petropavlovsk-Kamchatsky, Kamchatka, Russia" text "125.4 km (77.9 mi) WNW" text "Population: 181216" list_item text "Vilyuchinsk, Kamchatka, Russia"…GPT‑6.1 Terra still failed 29 tasks, 21 of them in multi-site data collection, and its most common miss is an empty cell rather than a wrong one. In 13 of the 29 it marked a value as not shown or unverifiable when the task expected one, and in 6 it filled in a wrong value. Nine more failed on a single site's figures, and on one task both models used all 100 steps without an answer. On the Polestar 3 specifications page, for example, it captured "250 kW DC" in the Charging section, decided the figure described a test charger and reported peak charging power as not shown.
GPT‑6.1 Terra reads 250 kW DC, then reports peak charging as not shown
ego-browser nodejs <<'EOF'const t=await taskSpace('eff4-high-ev-specs-01-…'),p=t.page('p1');await p.goto('https://www.polestar.com/us/polestar-3/specifications/');console.log((await p.evaluate(()=>document.body.innerText)).slice(0,25000));EOF…Charging³Battery capacity111 kWh250 kW DC10–80% in 30 min11 kW AC…Speed
On the 25 tasks both models passed, GPT‑6.1 Terra was faster on average: 200.9 seconds per task against 233.0, with medians of 161.5 and 194.1 seconds. It was faster on 17 of the 25. The whole lead comes from web apps automation, which makes up 14 of the 25 shared tasks.
Average time per task, lower is better
In web apps automation GPT‑6.1 Terra averaged 139.8 seconds against 225.6 on the 14 tasks both passed, and was faster on 12. On the other 11 shared tasks it was faster on 5, and GPT‑5.6 Terra had the lower average, 242.5 seconds against 278.7; one social media verification task alone took GPT‑6.1 Terra 947 seconds. The two models were served through different providers, so part of any gap may come from the endpoint rather than the model. The breakdown is in How we measured.
Cost
GPT‑6.1 Terra cost $0.286 per task against $0.742 for GPT‑5.6 Terra, 61% less, and it was cheaper in every type of work. On the 25 tasks both models passed, it cost $0.146 against $0.416. Divided by the tasks each model passed, that's $0.479 per passed task against $1.979.
Cost per task, by type of work
The whole saving is price. OpenAI lists GPT‑6.1 Terra at $1.00 per million input tokens and $5.00 per million output tokens, against $2.00 and $12.00 for GPT‑5.6 Terra. GPT‑6.1 Terra didn't use fewer tokens. It took more agent turns per task, 40.0 against 34.2, and read 31% more input tokens.
Cache reads are the largest part of both bills, because the agent re-sends the whole conversation on every step. Per task, GPT‑6.1 Terra read 2.66 million cached input tokens against 1.99 million, and wrote 6,752 output tokens against 9,788.
Multi-site data collection
If your agent collects the same data from many official sites, GPT‑6.1 Terra is the model to use. It passed 11 of our 32 multi-site data collection tasks, against 2 of 32 for GPT‑5.6 Terra, at $0.416 per task against $1.010. Each of these tasks gathers the same fields from eight or more sites into one table, and passes only if every field is right.
Our nine-browser release tracker shows how. Both models went through each vendor's release notes for its newest stable desktop version and security-fix count. GPT‑5.6 Terra copied an upstream Mozilla advisory into Waterfox's row and overcounted the total. GPT‑6.1 Terra noticed the page never claims those fixes as Waterfox's own and left the cells empty, as the answer key does. This is still hard work for both models: GPT‑6.1 Terra failed 21 of the 32 tasks, mostly by leaving a value empty, so check the empty cells before you use the table. The one multi-site task it lost was a week of number-one singles on national charts: the UK chart page returned a 403 to GPT‑6.1 Terra and it left those values out, while GPT‑5.6 Terra got every item right.
Web apps automation
For web apps automation, GPT‑6.1 Terra is faster as well as cheaper. Our 15 tasks run in test apps for operations, travel expenses and a cloud console, which generate fresh data for each attempt. GPT‑6.1 Terra passed all 15 against 14, at $0.096 per task against $0.331, and averaged 139.8 seconds against 225.6 on the 14 both passed. The one task it gained asked the agent to bring tier-1 monitors up to a written standard: GPT‑5.6 Terra picked the wrong auth-service monitor, and GPT‑6.1 Terra edited the right one.
Research and synthesis
Research and synthesis is where switching changes least. These 12 tasks give the agent a goal and a few sites and leave the path to it. GPT‑6.1 Terra passed 5 against 4, at $0.201 per task against $0.538, and was faster on 2 of the 3 tasks both passed. It is also one of two types of work where it lost a task. Asked which GLM 5.3 provider endpoint to pin on OpenRouter, it copied a provider's uptime from the recent-uptime chart (100.00%) instead of the uptime field it had read (99.91%). GPT‑5.6 Terra reported that field correctly, and with either model we'd keep a person on the final call for this kind of work.
Which model to use
If your agent collects data from many sites into one table, switch to GPT‑6.1 Terra: it passed 11 multi-site tasks to GPT‑5.6 Terra's 2, at well under half the cost. If your agent automates web apps, switch too, since GPT‑6.1 Terra was faster there as well as cheaper. For search and compare tasks and research and synthesis it still passed more tasks and cost less, with a smaller lead.
In web apps automation, switching also saves time: GPT‑6.1 Terra was faster on 12 of the 14 tasks both models passed. For other work the comparison rests on 11 shared tasks, where GPT‑5.6 Terra was faster on 6 and had the lower average, 242.5 seconds against 278.7. If speed matters for that work, time both models on your own tasks.
To get these results, run GPT‑6.1 Terra (gpt-6.1-terra) at high reasoning effort in the pi agent harness, with its bash tool and a skill file for ego-browser. ego-browser comes with ego (lite) and drives it as the agent's browser.
FAQ
Is GPT‑6.1 Terra better than GPT‑5.6 Terra for browser agents?
Yes, on our benchmark. GPT‑6.1 Terra passed 59.7% of tasks (43 of 72) against 37.5% (27 of 72) for GPT‑5.6 Terra. Both ran the same 72 web tasks in ego-browser under the same agent harness and step budget. The gain came mostly from multi-site data collection, 11 of 32 against 2 of 32. The full setup is in How we tested.
Is GPT‑6.1 Terra cheaper than GPT‑5.6 Terra per browser task?
Yes, by more than half. In our test of 72 browser tasks, GPT‑6.1 Terra cost $0.286 per task and GPT‑5.6 Terra $0.742, and it was cheaper in every type of work. On the 25 tasks both passed, it was $0.146 against $0.416. The saving comes from OpenAI's lower list prices, since GPT‑6.1 Terra read 31% more input tokens per task. The full setup is in How we tested.
Is GPT‑6.1 Terra faster than GPT‑5.6 Terra?
On average, yes, on the tasks both models passed. On the 25 tasks both models passed in our 72-task benchmark, GPT‑6.1 Terra took 200.9 seconds per task and GPT‑5.6 Terra 233.0 seconds, and GPT‑6.1 Terra was faster on 17. The lead comes from web apps automation, 139.8 seconds against 225.6. On the other 11 shared tasks GPT‑5.6 Terra had the lower average, 242.5 seconds against 278.7. The full setup is in How we tested.
How much does GPT‑6.1 Terra cost per million tokens?
OpenAI lists GPT‑6.1 Terra at $1.00 per million input tokens, $0.05 cached input, $1.25 for cache writes and $5.00 output. GPT‑5.6 Terra is listed at $2.00 input, $0.20 cached input and $12.00 output. In our 72-task browser benchmark that worked out to $0.286 per task against $0.742. Cache reads were the largest part of both bills. The full setup is in How we tested.
Which browser tasks improved the most with GPT‑6.1 Terra?
Multi-site data collection improved the most. GPT‑6.1 Terra passed 11 of 32 tasks that collect the same fields from eight or more official sites, against 2 of 32 for GPT‑5.6 Terra. That's 10 of its 18 gains in our 72-task benchmark, run in ego-browser under the same harness. The fixes came from reading each site's exact field, such as the opening keynote rather than a later talk. The full setup is in How we tested.
Did GPT‑6.1 Terra get worse on any browser task?
On pass or fail, on two tasks out of 72. GPT‑6.1 Terra passed 18 tasks GPT‑5.6 Terra failed and failed two GPT‑5.6 Terra passed. On an OpenRouter provider-uptime check it copied a recent-uptime chart, reporting a provider at 100.00% where the uptime field showed 99.91%. On a week of national singles charts, the UK chart page returned a 403 and it left those values out. By scored items it met fewer than GPT‑5.6 Terra on 10 tasks, 8 of which both failed. The full setup is in How we tested.
What does GPT‑6.1 Terra still get wrong in a browser?
Most often it leaves a value empty that the task expects. Of its 29 failed tasks in our 72-task benchmark, 13 had a value marked not shown or unverifiable, and 6 had a wrong value. On the 18 tasks GPT‑6.1 Terra gained, GPT‑5.6 Terra had written a wrong value in 14. The full setup is in How we tested.
How does GPT‑6.1 Terra handle web apps automation?
It passed all 15 web apps automation tasks in our benchmark, against 14 for GPT‑5.6 Terra. The tasks run in test apps for operations, travel expenses and a cloud console, with fresh data for each attempt. It cost $0.096 per task against $0.331, and took 139.8 seconds against 225.6 on the 14 tasks both passed. The full setup is in How we tested.
Should I switch my browser agent from GPT‑5.6 Terra to GPT‑6.1 Terra?
Yes, if your agent collects data across many sites or automates web apps. In our 72-task benchmark, GPT‑6.1 Terra passed 43 tasks against 27 and cost $0.286 per task against $0.742. On the 25 tasks both passed it was also faster on average, 200.9 seconds against 233.0, a lead that comes from web apps automation. The full setup is in How we tested.
What settings did you use to test GPT‑6.1 Terra?
We ran gpt-6.1-terra and gpt-5.6-terra at high reasoning effort through the pi agent harness, with ego-browser as the browser tool. Each model got the same 72 tasks, with a budget of 100 steps per task and up to two reruns for a task that failed. More detail is in How we tested, and every task is in Full results.
How we tested
Setup
We ran GPT‑6.1 Terra and GPT‑5.6 Terra through the same agent harness, pi, at high reasoning effort, with ego-browser as the browser tool. GPT‑6.1 Terra was served through the GitHub Copilot provider and GPT‑5.6 Terra through the OpenAI Codex provider, so part of any time difference may come from the endpoint. The model drives ego-browser by writing short JavaScript programs that load pages, click, fill forms and print page text or an accessibility snapshot. Each model ran each of the 72 tasks with a budget of 100 steps per task, and a task that failed could be rerun up to two more times.
Tasks
The 72 tasks cover five types of work. Multi-site data collection (32 tasks) gathers the same fields from eight or more official sites, such as nine seismology agencies' readings of one earthquake. Research and synthesis (12) gives a goal and a few sites, such as picking an OpenRouter provider endpoint. Search and compare (8) tasks compare figures across one to three sites, such as clinical trial registry entries against drug labels. Social media verification (5) matches posts on X against other records, and web apps automation (15) runs in test apps for operations, travel expenses and a cloud console. Each type comes from the task's own group in the dataset.
What counts as a pass
Each task has a list of scored items, 489 in all, including one per task that checks the whole deliverable. Most are values fixed in advance from the official sites, and the rest come from the business apps' data or are checked against the live page. A task passes only when its final answer meets every item.
How we measured
Time per task is the attempt's full wall-clock time: the model's thinking and responses, the browser's work, and the harness's own time between steps, which came to under two seconds per task. Speed is compared on the 25 tasks both models passed, since time spent on a failed task can't be compared. Cost is each attempt's recorded token usage at OpenAI's list prices. The two attempts that used all 100 steps count as failures.
| Tasks both models passed | Tasks | GPT‑6.1 Terra | GPT‑5.6 Terra | GPT‑6.1 Terra faster on |
|---|---|---|---|---|
| All, average | 25 | 200.9 s | 233.0 s | 17 of 25 |
| All, median | 25 | 161.5 s | 194.1 s | — |
| Web apps automation, average | 14 | 139.8 s | 225.6 s | 12 of 14 |
| Other types of work, average | 11 | 278.7 s | 242.5 s | 5 of 11 |
The 11 other shared tasks are one multi-site data collection task, three research and synthesis tasks, five search and compare tasks and two social media verification tasks, too few of each to compare separately. Web apps automation makes up 14 of the 25 shared tasks, against 15 of all 72, so the overall average leans toward it.
Full results
Every task, with each model's result and how many of its scored items the final answer met.
| Task | Type of work | GPT‑6.1 Terra | GPT‑5.6 Terra | Result |
|---|---|---|---|---|
| 2025 climate scorecard from eleven weather services | Multi-site data collection | Fail 0/7 | Fail 0/7 | Both failed |
| 2025 traffic at eight hub airports | Multi-site data collection | Fail 0/8 | Fail 1/8 | Both failed |
| 2025-26 season figures from nine football leagues | Multi-site data collection | Fail 0/8 | Fail 0/8 | Both failed |
| 2026 top prizes at nine film festivals | Multi-site data collection | Fail 5/8 | Fail 5/8 | Both failed |
| Approval sheet for nine open-weight models on Hugging Face | Multi-site data collection | Fail 5/8 | Fail 6/8 | Both failed |
| Compliance matrix from nine SaaS trust centers | Multi-site data collection | Fail 5/7 | Fail 5/7 | Both failed |
| Crash-test ratings for one SUV across nine programs | Multi-site data collection | Fail 1/7 | Fail 1/7 | Both failed |
| Editions in force for nine technical standards | Multi-site data collection | Pass 7/7 | Fail 2/7 | Only GPT‑6.1 Terra passed |
| First 2026 rate decision at twelve central banks | Multi-site data collection | Pass 7/7 | Fail 5/7 | Only GPT‑6.1 Terra passed |
| First and latest Q2 2026 GDP prints at nine agencies | Multi-site data collection | Pass 8/8 | Fail 4/8 | Only GPT‑6.1 Terra passed |
| Free CI plans for private repos at nine services | Multi-site data collection | Fail 5/7 | Fail 0/7 | Both failed |
| Free tiers of nine serverless platforms | Multi-site data collection | Pass 7/7 | Fail 1/7 | Only GPT‑6.1 Terra passed |
| How nine music services list one album | Multi-site data collection | Fail 3/7 | Fail 5/7 | Both failed |
| Latest election results from nine election authorities | Multi-site data collection | Fail 5/8 | Fail 1/8 | Both failed |
| Newest release line of nine language runtimes | Multi-site data collection | Fail 5/7 | Fail 3/7 | Both failed |
| Newest stable release of nine JavaScript frameworks | Multi-site data collection | Pass 6/6 | Pass 6/6 | Both passed |
| Newest stable release of nine desktop browsers | Multi-site data collection | Pass 6/6 | Fail 2/6 | Only GPT‑6.1 Terra passed |
| Nine agencies' readings of the 2025 Kamchatka earthquake | Multi-site data collection | Pass 7/7 | Fail 5/7 | Only GPT‑6.1 Terra passed |
| Official UK specs of nine flagship phones | Multi-site data collection | Pass 7/7 | Fail 5/7 | Only GPT‑6.1 Terra passed |
| Official specs of eight cameras | Multi-site data collection | Fail 1/7 | Fail 3/7 | Both failed |
| One book's edition details on ten publisher and library sites | Multi-site data collection | Fail 0/8 | Fail 0/8 | Both failed |
| One game's age ratings at eight rating boards | Multi-site data collection | Fail 0/7 | Fail 1/7 | Both failed |
| One sudo CVE's fix status in nine vendor advisories | Multi-site data collection | Fail 2/7 | Fail 2/7 | Both failed |
| One week's number-one singles on nine national charts | Multi-site data collection | Fail 0/8 | Pass 8/8 | Only GPT‑5.6 Terra passed |
| Opening keynotes of nine developer conferences | Multi-site data collection | Pass 7/7 | Fail 2/7 | Only GPT‑6.1 Terra passed |
| Protocol Buffers releases on nine package registries | Multi-site data collection | Fail 3/8 | Fail 4/8 | Both failed |
| Results from eight 2025 developer surveys | Multi-site data collection | Pass 7/7 | Fail 3/7 | Only GPT‑6.1 Terra passed |
| Statutory minimum wage in nine places | Multi-site data collection | Fail 2/7 | Fail 3/7 | Both failed |
| Support end dates for nine server products | Multi-site data collection | Fail 5/7 | Fail 3/7 | Both failed |
| Ten museums' records of Hokusai's Great Wave | Multi-site data collection | Fail 6/8 | Fail 1/8 | Both failed |
| Tirzepatide weight-management approvals at eight regulators | Multi-site data collection | Pass 7/7 | Fail 1/7 | Only GPT‑6.1 Terra passed |
| US range and charging figures for nine electric SUVs | Multi-site data collection | Fail 5/7 | Fail 5/7 | Both failed |
| Cheap reasoning settings on the ARC Prize leaderboards | Research and synthesis | Pass 6/6 | Pass 6/6 | Both passed |
| Count models above EU compute markers in Epoch AI's data | Research and synthesis | Fail 4/9 | Fail 4/9 | Both failed |
| EU AI Act disclosures for eight frontier models | Research and synthesis | Fail 6/11 | Fail 7/11 | Both failed |
| Every model launched in September 2026, from six API changelogs | Research and synthesis | Pass 7/7 | Pass 7/7 | Both passed |
| Fact-check ten claims about September 2026 model launches | Research and synthesis | Fail 1/5 | Fail 1/5 | Both failed |
| Pick a main and backup API among five Chinese models | Research and synthesis | Pass 6/6 | Pass 6/6 | Both passed |
| Pick a model offered in one Azure Germany region | Research and synthesis | Pass 7/7 | Fail 4/7 | Only GPT‑6.1 Terra passed |
| Pick the fastest GLM 5.3 provider endpoint on OpenRouter | Research and synthesis | Fail 8/10 | Pass 10/10 | Only GPT‑5.6 Terra passed |
| Price one API workload across six models | Research and synthesis | Fail 5/9 | Fail 3/9 | Both failed |
| Six providers' API prices as the Wayback Machine kept them | Research and synthesis | Fail 6/9 | Fail 6/9 | Both failed |
| Trace the sources behind a Wikipedia LLM table | Research and synthesis | Pass 8/8 | Fail 6/8 | Only GPT‑6.1 Terra passed |
| What one LiveBench release showed | Research and synthesis | Fail 5/7 | Fail 5/7 | Both failed |
| Build an inflation table from FRED and cross-check it on BLS | Search and compare | Pass 8/8 | Pass 8/8 | Both passed |
| Check a GPT lecture's numbers against the video, code and paper | Search and compare | Pass 6/6 | Fail 4/6 | Only GPT‑6.1 Terra passed |
| Compare three MacBook configurations on Apple's US store | Search and compare | Pass 9/9 | Pass 9/9 | Both passed |
| Cross-check Phase 3 trials on ClinicalTrials.gov against DailyMed labels | Search and compare | Pass 9/9 | Fail 6/9 | Only GPT‑6.1 Terra passed |
| Go/no-go list for six web features from MDN and Can I use | Search and compare | Pass 5/5 | Fail 3/5 | Only GPT‑6.1 Terra passed |
| Price a team offsite's flights to Denver | Search and compare | Pass 11/11 | Pass 11/11 | Both passed |
| Reconcile six Wes Anderson films' grosses between a box-office database and Wikipedia | Search and compare | Pass 9/9 | Pass 9/9 | Both passed |
| Register snapshot of four UK companies on Companies House | Search and compare | Pass 9/9 | Pass 9/9 | Both passed |
| Eight brands' 2026 product announcements on X and newsrooms | Social media verification | Pass 7/7 | Fail 4/7 | Only GPT‑6.1 Terra passed |
| Eight football clubs' summer signings on X and club sites | Social media verification | Fail 0/7 | Fail 0/7 | Both failed |
| Eight open-source projects' security releases on X and their sites | Social media verification | Pass 7/7 | Fail 5/7 | Only GPT‑6.1 Terra passed |
| How fast eight universities posted their 2025 Nobel news on X | Social media verification | Pass 7/7 | Pass 7/7 | Both passed |
| Match SpaceX's launch posts on X to Wikipedia's launch lists | Social media verification | Pass 5/5 | Pass 5/5 | Both passed |
| Apply a lockbox deposit to open invoices | Web apps automation | Pass 4/4 | Pass 4/4 | Both passed |
| Audit and roll back changes made during a change freeze | Web apps automation | Pass 4/4 | Pass 4/4 | Both passed |
| Book a colleague's trip in a travel app | Web apps automation | Pass 5/5 | Pass 5/5 | Both passed |
| Bring tier-1 monitors up to a written standard | Web apps automation | Pass 4/4 | Fail 1/4 | Only GPT‑6.1 Terra passed |
| Claim a week of client-visit mileage | Web apps automation | Pass 4/4 | Pass 4/4 | Both passed |
| Close February receivables and open disputes in an operations app | Web apps automation | Pass 5/5 | Pass 5/5 | Both passed |
| File a two-city per diem claim | Web apps automation | Pass 4/4 | Pass 4/4 | Both passed |
| File an incident postmortem timeline in a cloud console | Web apps automation | Pass 4/4 | Pass 4/4 | Both passed |
| File damage claims for two customers' deliveries | Web apps automation | Pass 4/4 | Pass 4/4 | Both passed |
| Fix next week's on-call schedule | Web apps automation | Pass 4/4 | Pass 4/4 | Both passed |
| Match card transactions to a draft expense report | Web apps automation | Pass 4/4 | Pass 4/4 | Both passed |
| Place a warehouse's replenishment orders | Web apps automation | Pass 4/4 | Pass 4/4 | Both passed |
| Review an employee's expense report | Web apps automation | Pass 4/4 | Pass 4/4 | Both passed |
| Set deploy freezes from error-budget burn | Web apps automation | Pass 4/4 | Pass 4/4 | Both passed |
| Settle carrier on-time figures and fix alert rules | Web apps automation | Pass 6/6 | Pass 6/6 | Both passed |
