Suites / flow

Flow

Salesforce Flow engineering: record-triggered flows (before-save fast field updates, after-save related-record automation, before-delete), entry conditions and $Record__Prior, decisions and formulas, Custom Error, fault paths, subflows, autolaunched flows called from Apex, invocable Apex actions, scheduled paths, and choosing between flows and Apex triggers. Answers are Flow metadata files (API 67.0) deployed check-only to a clean scratch org, where hidden Apex tests do DML (including 200-record bulk tests that assert on Limits) and check the outcome.

18
tasks
Execution
grading
100%
highest c@3 · Claude Fable 5.1, joint first with 15 others
78
configurations with results

How it's graded

Flow metadata is deployed as a check-only validation to a clean scratch org, where hidden Apex tests do DML (including 200-record bulk tests that assert on governor limits) and check the outcome. Decision-guide questions (flow or Apex?) are multiple choice.

Difficulty mix

5 easy7 medium6 hard

Leaderboard

Who's best at Flow

Within three attempts (c@3) on this suite only, with 95% confidence intervals. With 18 tasks the intervals are wide: where two overlap, those configurations are not clearly different. 1 finished configuration isn’t shown yet: its retries haven’t finished. Open in the full leaderboard →

API: the best of each lab at Flow
Within three attempts (c@3) on Flow, on the API board (models reached through a hosted service). Bars whose intervals overlap are not clearly different.
  • 95% interval
=1Gemini 4 Argon low tie
100%, 95% interval 82–100%
=1GPT-6 Astra medium tie
100%, 95% interval 82–100%
=1Claude Opus 4.8 thinking off tie
100%, 95% interval 82–100%
=1Grok 4.6 high tie
100%, 95% interval 82–100%
5Muse Wonder high
94%, 95% interval 74–99%
6DeepSeek V4.1 Flash high
61%, 95% interval 39–80%
7GLM-5.3 low
50%, 95% interval 29–71%
8Inkling Small high
44%, 95% interval 25–66%
=9MiMo V2.6 Pro thinking on tie
28%, 95% interval 13–51%
=9Mistral Large 4 high tie
28%, 95% interval 13–51%
11Nemotron 3 Ultra thinking on
22%, 95% interval 9–45%
12Laguna S 2.1 thinking on
17%, 95% interval 6–39%
=13Ling 3.0 Flash thinking on tie
11%, 95% interval 3–33%
=13Qwen3.8 Flash-Next thinking on tie
11%, 95% interval 3–33%
    Show the numbers
    ModelScore95% CI
    Gemini 4 Argon low100%82–100
    GPT-6 Astra medium100%82–100
    Claude Opus 4.8 thinking off100%82–100
    Grok 4.6 high100%82–100
    Muse Wonder high94%74–99
    DeepSeek V4.1 Flash high61%39–80
    GLM-5.3 low50%29–71
    Inkling Small high44%25–66
    MiMo V2.6 Pro thinking on28%13–51
    Mistral Large 4 high28%13–51
    Nemotron 3 Ultra thinking on22%9–45
    Laguna S 2.1 thinking on17%6–39
    Ling 3.0 Flash thinking on11%3–33
    Qwen3.8 Flash-Next thinking on11%3–33
    Local: the best of each lab at Flow
    Within three attempts (c@3) on Flow, on the Local board (models served on Forcebench’s own hardware). Bars whose intervals overlap are not clearly different.
    • 95% interval
    1DeepSeek V4.1 Flash 1017 high · Native FP8, FP4 experts
    72%, 95% interval 49–88%
    2GLM-5.3 Flash low · EXL3 4.0bpw experts, FP8 dense
    56%, 95% interval 34–75%
    3Qwen3.8 Flash-Next medium · NVFP4
    28%, 95% interval 13–51%
    =4Muse Glimmer 30B high · NVFP4 + FP8 mixed (NVIDIA ModelOpt) tie
    22%, 95% interval 9–45%
    =4Nemotron 3.5 Super VL thinking on · NVFP4 mixed precision (NVIDIA ModelOpt) tie
    22%, 95% interval 9–45%
    =6Gemma 4 26B-A4B thinking on · NVFP4 tie
    17%, 95% interval 6–39%
    =6MiniMax M3.1 Flash high · NVFP4 experts, MXFP8 (QAT) tie
    17%, 95% interval 6–39%
    =6MiMo V2.6 Flash thinking on · FP8 + MXFP4 experts tie
    17%, 95% interval 6–39%
    =9MiniCPM-V 4.7 35B-A3B thinking on · BF16 tie
    11%, 95% interval 3–33%
    =9Ornith 1.5 122B thinking on · NVFP4 experts, FP8 attention (NVIDIA ModelOpt) tie
    11%, 95% interval 3–33%
      Show the numbers
      ModelScore95% CI
      DeepSeek V4.1 Flash 1017 high · Native FP8, FP4 experts72%49–88
      GLM-5.3 Flash low · EXL3 4.0bpw experts, FP8 dense56%34–75
      Qwen3.8 Flash-Next medium · NVFP428%13–51
      Muse Glimmer 30B high · NVFP4 + FP8 mixed (NVIDIA ModelOpt)22%9–45
      Nemotron 3.5 Super VL thinking on · NVFP4 mixed precision (NVIDIA ModelOpt)22%9–45
      Gemma 4 26B-A4B thinking on · NVFP417%6–39
      MiniMax M3.1 Flash high · NVFP4 experts, MXFP8 (QAT)17%6–39
      MiMo V2.6 Flash thinking on · FP8 + MXFP4 experts17%6–39
      MiniCPM-V 4.7 35B-A3B thinking on · BF1611%3–33
      Ornith 1.5 122B thinking on · NVFP4 experts, FP8 attention (NVIDIA ModelOpt)11%3–33
      Flow: c@3 by configuration with 95% confidence intervals
      #Model and configScore c@3 · 95% CITasks
      1Claude Fable 5.1 effort high
      100%95% CI 82–100
      		18
      1Claude Opus 4.8 effort high
      100%95% CI 82–100
      		18
      1Claude Opus 4.8 effort medium
      100%95% CI 82–100
      		18
      1Claude Opus 4.8 effort low
      100%95% CI 82–100
      		18
      1Claude Opus 4.8 thinking off
      100%95% CI 82–100
      		18
      1Claude Opus 5.5 effort medium
      100%95% CI 82–100
      		18
      1Claude Opus 5.5 effort low
      100%95% CI 82–100
      		18
      1Claude Sonnet 5 effort high
      100%95% CI 82–100
      		18
      1Claude Sonnet 5.5 effort high
      100%95% CI 82–100
      		18
      1Gemini 4 Argon effort low
      100%95% CI 82–100
      		18
      1GPT-5.6 Sol effort medium
      100%95% CI 82–100
      		18
      1GPT-6 Astra effort medium
      100%95% CI 82–100
      		18
      1GPT-6 Astra effort high
      100%95% CI 82–100
      		18
      1GPT-6.1 Sol effort high
      100%95% CI 82–100
      		18
      1Grok 4.6 effort high
      100%95% CI 82–100
      		18
      1Grok 4.7 effort high
      100%95% CI 82–100
      		18
      17Claude Haiku 5.5 effort high
      94%95% CI 74–99
      		18
      17Claude Opus 5.5 effort high
      94%95% CI 74–99
      		18
      17Claude Sonnet 5 effort medium
      94%95% CI 74–99
      		18
      17Claude Sonnet 5.5 effort low
      94%95% CI 74–99
      		18
      17Claude Sonnet 5.5 effort medium
      94%95% CI 74–99
      		18
      17Gemini 4 Argon effort medium
      94%95% CI 74–99
      		18
      17Gemini 4 Argon effort high
      94%95% CI 74–99
      		18
      17GPT-6 Luna effort medium
      94%95% CI 74–99
      		18
      17Muse Wonder effort high
      94%95% CI 74–99
      		18
      26Claude Sonnet 5 thinking off
      89%95% CI 67–97
      		18
      26Gemini 3.8 Flash effort medium
      89%95% CI 67–97
      		18
      28Muse Wonder effort low
      83%95% CI 61–94
      		18
      29Claude Haiku 5.5 effort medium
      78%95% CI 55–91
      		18
      29Claude Haiku 5.5 effort low
      78%95% CI 55–91
      		18
      29Claude Sonnet 5 effort low
      78%95% CI 55–91
      		18
      29Muse Wonder effort medium
      78%95% CI 55–91
      		18
      29Muse Wonder effort max
      78%95% CI 55–91
      		18
      34DeepSeek V4.1 Flash 1017 Native FP8, FP4 experts · SGLang · effort high
      72%95% CI 49–88
      		18
      35DeepSeek V4.1 Flash FP8 · effort high
      61%95% CI 39–80
      		18
      35DeepSeek V4.1 Flash EXL3 2.9bpw · vLLM + ExLlamaV3 · effort high
      61%95% CI 39–80
      		18
      37DeepSeek V4.1 Flash Native FP8, FP4 experts · SGLang · effort high
      56%95% CI 34–75
      		18
      37GLM-5.3 Flash EXL3 4.0bpw experts, FP8 dense · vLLM + ExLlamaV3 · effort low
      56%95% CI 34–75
      		18
      39GLM-5.3 FP8 · effort low
      50%95% CI 29–71
      		18
      40Inkling Small FP8 · effort high
      44%95% CI 25–66
      		18
      41GLM-5.3 Flash EXL3 4.0bpw experts, FP8 dense · vLLM + ExLlamaV3 · effort high
      33%95% CI 16–56
      		18
      42GLM-5.3 Flash EXL3 4.0bpw experts, 4-bit dense · TensorFold · effort high
      28%95% CI 13–51
      		18
      42MiMo V2.6 Pro FP8 · thinking on
      28%95% CI 13–51
      		18
      42Mistral Large 4 effort high
      28%95% CI 13–51
      		18
      42Qwen3.8 Flash-Next NVFP4 · vLLM · effort medium
      28%95% CI 13–51
      		18
      46GLM-5.3 Flash FP8 · effort high
      22%95% CI 9–45
      		18
      46GLM-5.3 Flash EXL3 4.0bpw experts, FP8 dense · vLLM + ExLlamaV3 · thinking off
      22%95% CI 9–45
      		18
      46Muse Glimmer 30B NVFP4 + FP8 mixed (NVIDIA ModelOpt) · vLLM · effort high
      22%95% CI 9–45
      		18
      46Nemotron 3 Ultra thinking on
      22%95% CI 9–45
      		18
      46Nemotron 3.5 Super VL NVFP4 mixed precision (NVIDIA ModelOpt) · TensorFold (Zig) · thinking on
      22%95% CI 9–45
      		18
      46Qwen3.8 Flash-Next MLX 4-bit · MTPLX · effort medium
      22%95% CI 9–45
      		18
      52Qwen3.8 27B AWQ-INT4 · vLLM · effort medium
      19%95% CI 7–42
      		18
      53Claude Haiku 4.5 thinking on
      17%95% CI 6–39
      		18
      53Claude Haiku 4.5 thinking off
      17%95% CI 6–39
      		18
      53DeepSeek V4 Flash Vision experimental FP8 + FP4 experts · vLLM · effort high
      17%95% CI 6–39
      		18
      53DeepSeek V4 Pro FP8 · effort high
      17%95% CI 6–39
      		18
      53Gemma 4 26B-A4B NVFP4 · vLLM · thinking on
      17%95% CI 6–39
      		18
      53Laguna S 2.1 FP4 · thinking on
      17%95% CI 6–39
      		18
      53MiMo V2.6 Flash FP8 + MXFP4 experts · SGLang · thinking on
      17%95% CI 6–39
      		18
      53MiMo V2.6 Flash FP8 · thinking on
      17%95% CI 6–39
      		18
      53MiniMax M3.1 Flash NVFP4 experts, MXFP8 (QAT) · vLLM · effort high
      17%95% CI 6–39
      		18
      53Qwen3.8 27B MLX 8-bit · MTPLX · effort low
      17%95% CI 6–39
      		18
      53Qwen3.8 27B MLX 4-bit · MTPLX · effort low
      17%95% CI 6–39
      		18
      53Qwen3.8 Flash-Next MLX 4-bit · TensorFold · effort medium
      17%95% CI 6–39
      		18
      53Qwen3.8 Flash-Next INT4-AutoRound, 5 of 10 experts · TensorFold (Zig) · effort medium
      17%95% CI 6–39
      		18
      53Qwen3.8 Flash-Next INT4-AutoRound, 5 of 10 experts · TensorFold (Zig) · effort medium
      17%95% CI 6–39
      		18
      67Gemma 4 31B QAT W4A16 · vLLM · thinking on
      11%95% CI 3–33
      		18
      67Ling 3.0 Flash BF16 · thinking on
      11%95% CI 3–33
      		18
      67MiniCPM-V 4.7 35B-A3B BF16 · vLLM · thinking on
      11%95% CI 3–33
      		18
      67MiniCPM-V 4.7 35B-A3B INT4 W4A16 experts, BF16 dense · vLLM · thinking on
      11%95% CI 3–33
      		18
      67Nemotron 3.5 Bolt IQ2_XXS + NVFP4 + Q8_0 mixed (NVIDIA ModelOpt) · TensorFold (Zig) · thinking on
      11%95% CI 3–33
      		18
      67Nemotron 3.5 Lightning NVFP4 · thinking on
      11%95% CI 3–33
      		18
      67Ornith 1.5 122B NVFP4 experts, FP8 attention (NVIDIA ModelOpt) · vLLM · thinking on
      11%95% CI 3–33
      		18
      67Qwen3.6 35B-A3B NVFP4 · vLLM · thinking on
      11%95% CI 3–33
      		18
      67Qwen3.8 27B AWQ-INT4 · vLLM · effort low
      11%95% CI 3–33
      		18
      67Qwen3.8 27B Splash 4-bit · Splash · effort low
      11%95% CI 3–33
      		18
      67Qwen3.8 27B AWQ-INT4 · vLLM · effort xhigh
      11%95% CI 3–33
      		18
      67Qwen3.8 Flash-Next thinking on
      11%95% CI 3–33
      		18

      Tasks

      What's in the suite

      Solve rate is the share of finished runs whose first answer to the task passed (pass@1), a rough guide to how hard models find it. Difficulty is the task author’s label; from results is what the solve rate says: easy when two thirds or more of the finished runs pass it, hard at a third or fewer, once five have graded it.

      Flow tasks with difficulty and solve rate
      TaskDifficultyFrom resultsSolve rate
      Autolaunched shipping quote flow called from Apexflow-shipping-quote-autolaunchedeasymedium
      38%
      Choose the tool for a reminder relative to a record dateflow-choice-close-date-remindereasyeasy
      100%
      Choose the tool for a same-record field updateflow-choice-same-record-updateeasyeasy
      100%
      Escalate breakdown cases with a fast field updateflow-case-breakdown-escalationeasymedium
      37%
      Welcome-call task for new customer accountsflow-account-customer-welcome-taskeasymedium
      35%
      Automation when a record is restored from the Recycle Binflow-choice-undelete-restoremediumeasy
      84%
      Block closing a case that still has open tasksflow-case-close-open-tasksmediumhard
      30%
      Call an invocable Apex credit check from a record-triggered flowflow-account-credit-check-actionmediumhard
      23%
      Close-date reminder on a scheduled pathflow-opportunity-close-reminder-scheduledmediumhard
      28%
      Create a delivery project when an opportunity is wonflow-opportunity-won-projectmediumhard
      32%
      Flag large opportunity amount swings using $Record__Priorflow-opportunity-amount-swingmediummedium
      38%
      Set a case SLA due date/time from its priorityflow-case-sla-duemediumhard
      20%
      Cascade an account discount to its open opportunities in bulkflow-account-discount-cascadehardhard
      20%
      Fault path that logs failed shipment creationflow-order-shipment-fault-pathhardhard
      29%
      Fix a flow that syncs account addresses to contactsflow-fix-account-address-synchardhard
      27%
      Invocable Apex contact matcher orchestrated by a flowflow-web-signup-contact-matchhardhard
      27%
      Record-Triggered Automation decision guide principlesflow-choice-guide-statementshardmedium
      41%
      Reusable onboarding subflow called from a record-triggered flowflow-account-onboarding-subflowhardhard
      24%

      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