Models / gpt-6-astra

GPT-6 Astra

OpenAI's model, with closed weights. Forcebench has run it in 2 configurations on the API board. API is models reached through a hosted service; Local, models served on Forcebench’s own hardware.

On the API board

Models reached through a hosted service. GPT-6 Astra is ranked by the highest-scoring of its 2 configurations: the second-highest score of 26 models, not clearly different from 5 other models. Scores are within 3 attempts (c@3) with 95% intervals; each configuration is compared with the highest-scoring one by whether their 95% intervals overlap. Every configuration on the board → Cost and speed →

95%
within 3 attempts (c@3), 95% CI 92–97 · medium · second-highest score of 26 models, not clearly different from 5 other models
846
output tokens per task · median 502, p90 2k
7.4 s
median per task, every attempt counted, including network and queueing · 90th percentile 28 s
0%
no answer within the 32,768-token budget
$0.031
per task at the list price of 10 October 2026, every attempt counted
$0.032
per solved task: the total cost over the tasks solved · cost and speed
Every configuration
Within 3 attempts (c@3) with 95% intervals, highest first. Bars whose intervals overlap are not clearly different.
  • 95% interval
medium OpenRouterhighest score
95%, 95% interval 92–97%
high OpenRouter
92%, 95% interval 90–95%
    Show the numbers
    Every configuration of GPT-6 Astra on the API board
    ConfigurationRank all configsWithin 3 attempts c@3 · 95% CIAgainst the highest by 95% interval overlapTokens mean · median · p90No answerLatency median per task, every attempt countedCost per taskCost per solved task
    OpenRouter · mediumhighest score2
    95%95% CI 92–97
    		—846502 · 2k0%7.4 s$0.031$0.032
    OpenRouter · high7
    92%95% CI 90–95
    		not clearly different1.8k865 · 4.6k0%12 s$0.061$0.065
    Suites, at its highest-scoring configuration
    medium · OpenRouter: within 3 attempts (c@3) on each suite with 95% intervals, against the median of the API board's models, each at its highest-scoring configuration.
    • 95% interval
    • API board median
    Apex
    100%, 95% interval 84–100%
    Salesforce APIs
    89%, 95% interval 67–97%
    CI/CD
    100%, 95% interval 82–100%
    Salesforce CLI
    100%, 95% interval 84–100%
    Salesforce docs
    82%, 95% interval 59–94%
    fflib Enterprise Patterns
    94%, 95% interval 74–99%
    Flow
    100%, 95% interval 82–100%
    Governor limits & pushback
    100%, 95% interval 83–100%
    Lightning Web Components
    100%, 95% interval 82–100%
    Nonprofit Success Pack
    94%, 95% interval 74–99%
    Packaging
    100%, 95% interval 82–100%
    Permissions & access
    100%, 95% interval 80–100%
    Scratch org definitions
    67%, 95% interval 44–84%
    SOQL
    100%, 95% interval 84–100%
    Trigger Actions Framework
    94%, 95% interval 74–99%
      Show the numbers
      Per-suite scores of GPT-6 Astra's highest-scoring configuration on the API board, with the board's median
      Suitec@3 95% CIAPI medianTasks
      Apex
      100%95% CI 84–100
      		90%20
      Salesforce APIs
      89%95% CI 67–97
      		72%18
      CI/CD
      100%95% CI 82–100
      		88%17
      Salesforce CLI
      100%95% CI 84–100
      		90%20
      Salesforce docs
      82%95% CI 59–94
      		29%17
      fflib Enterprise Patterns
      94%95% CI 74–99
      		61%18
      Flow
      100%95% CI 82–100
      		69%18
      Governor limits & pushback
      100%95% CI 83–100
      		89%19
      Lightning Web Components
      100%95% CI 82–100
      		100%18
      Nonprofit Success Pack
      94%95% CI 74–99
      		81%18
      Packaging
      100%95% CI 82–100
      		69%18
      Permissions & access
      100%95% CI 80–100
      		87%15
      Scratch org definitions
      67%95% CI 44–84
      		50%18
      SOQL
      100%95% CI 84–100
      		100%20
      Trigger Actions Framework
      94%95% CI 74–99
      		86%18

      By reasoning effort

      The same configuration at several efforts, each against its lowest reasoning effort on the same tasks (a paired 95% interval), and charted against the effort tier and against the output tokens it costs. These scores are pass@1, the first answer alone: a paired comparison needs both efforts' answers to the same tasks, and not every effort has had its retries. Lite-subset sweeps (4 tasks per suite) are never plotted with the full benchmark. The effort study has every model.

      GPT-6 Astra by reasoning effort, full benchmark: each effort against the sweep's lowest reasoning effort
      Configuration · effort · full benchmarkpass@1 95% CIAgainst the lowest reasoning effort same tasks, pairedTokens
      mediumbaseline
      84%95% CI 80–87
      		—846
      high
      80%95% CI 76–84
      		clearly lower (−4 pts, 95% CI −7 pts to −0.1 pts)1.8k
      Score by effort tier · full benchmark
      Overall pass@1; vertical bars are 95% confidence intervals.
      GPT-6 Astra, full benchmark: pass@1 by reasoning effort tierGPT-6 Astra: medium 84%, high 80%0%25%50%75%100%MediumHighReasoning effort tierpass@1GPT-6 Astra
      Show the numbers
      ModelEffort (tier)pass@195% CITasksNo answerTokens: meanmedianReasoning tokensLatency: median per answer
      GPT-6 Astramedium84%80–872720%8465024136.9 s
      GPT-6 Astrahigh80%76–842720%1.8k8651.4k12 s
      Score against output tokens · full benchmark
      Each point is one effort setting. Further right means more output tokens per task. The numbers are under the chart above.
      GPT-6 Astra, full benchmark: pass@1 against mean output tokens per taskGPT-6 Astra: 846 tokens 84%, 1.8k tokens 80%0%25%50%75%100%1k2kMean output tokens per task (log scale)pass@1mediumhighGPT-6 Astra

      In the studies

      Private evaluations

      Want this for your team?

      Azul Labs runs Forcebench privately for two kinds of Salesforce team.

      Choosing a model to run yourself?

      A private benchmark of your shortlist

      Open-weight models at the quantisations and reasoning efforts you could run on infrastructure you control, tested on your kind of Salesforce work, with a written recommendation.

      • Private tasks written from your work only ever run on local models; they are never sent to an API.
      • Re-testing after any change of model, quantisation or engine is available.
      • Typical turnaround is 2–3 weeks; large models take about a day of machine time per configuration.
      Get a private benchmarkHow it works

      Building or buying a Salesforce AI tool?

      A private evaluation of your tool

      Your tool on held-out tasks it has never seen, against API baselines (models reached through a hosted service) and Local ones (served on Forcebench’s own hardware), with a written report on where it holds up and where it fails.

      • It can start as a smaller pilot, about 30 held-out tasks with your tool against two baselines, before a full evaluation.
      • Repeat evaluations after your releases are available.
      • Typical turnaround is 2–3 weeks.
      Ask about an evaluationHow it works

      Fixed fee, quoted up front. The quote depends on the number of tasks and the number of models, tools or configurations tested.

      Or email leo@azl.au · azl.au