Models / claude-haiku-4-5

Claude Haiku 4.5

Anthropic'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. Claude Haiku 4.5 is ranked by the highest-scoring of its 2 configurations: the 22nd-highest score of 26 models, not clearly different from 3 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 →

50%
within 3 attempts (c@3), 95% CI 45–55 · thinking on · 22nd-highest score of 26 models, not clearly different from 3 other models
5.1k
output tokens per task · median 4.6k, p90 10k
41 s
median per task, every attempt counted, including network and queueing · 90th percentile 154 s
0%
no answer within the 32,768-token budget
$0.044
per task at the list price of 10 October 2026, every attempt counted
$0.089
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
thinking on Anthropichighest score
50%, 95% interval 45–55%
thinking off Anthropic
46%, 95% interval 41–52%
    Show the numbers
    Every configuration of Claude Haiku 4.5 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
    Anthropic · thinking onhighest score35
    50%95% CI 45–55
    		—5.1k4.6k · 10k0%41 s$0.044$0.089
    Anthropic · thinking off36
    46%95% CI 41–52
    		not clearly different1.1k657 · 2.3k0%8.0 s$0.015$0.032
    Suites, at its highest-scoring configuration
    thinking on · Anthropic: 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
    75%, 95% interval 53–89%
    Salesforce APIs
    44%, 95% interval 25–66%
    CI/CD
    65%, 95% interval 41–83%
    Salesforce CLI
    60%, 95% interval 39–78%
    Salesforce docs
    0%, 95% interval 0–18%
    fflib Enterprise Patterns
    44%, 95% interval 25–66%
    Flow
    17%, 95% interval 6–39%
    Governor limits & pushback
    79%, 95% interval 57–91%
    Lightning Web Components
    89%, 95% interval 67–97%
    Nonprofit Success Pack
    56%, 95% interval 34–75%
    Packaging
    33%, 95% interval 16–56%
    Permissions & access
    73%, 95% interval 48–89%
    Scratch org definitions
    11%, 95% interval 3–33%
    SOQL
    70%, 95% interval 48–85%
    Trigger Actions Framework
    28%, 95% interval 13–51%
      Show the numbers
      Per-suite scores of Claude Haiku 4.5's highest-scoring configuration on the API board, with the board's median
      Suitec@3 95% CIAPI medianTasks
      Apex
      75%95% CI 53–89
      		90%20
      Salesforce APIs
      44%95% CI 25–66
      		72%18
      CI/CD
      65%95% CI 41–83
      		88%17
      Salesforce CLI
      60%95% CI 39–78
      		90%20
      Salesforce docs
      0%95% CI 0–18
      		29%17
      fflib Enterprise Patterns
      44%95% CI 25–66
      		61%18
      Flow
      17%95% CI 6–39
      		69%18
      Governor limits & pushback
      79%95% CI 57–91
      		89%19
      Lightning Web Components
      89%95% CI 67–97
      		100%18
      Nonprofit Success Pack
      56%95% CI 34–75
      		81%18
      Packaging
      33%95% CI 16–56
      		69%18
      Permissions & access
      73%95% CI 48–89
      		87%15
      Scratch org definitions
      11%95% CI 3–33
      		50%18
      SOQL
      70%95% CI 48–85
      		100%20
      Trigger Actions Framework
      28%95% CI 13–51
      		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.

      Claude Haiku 4.5 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
      thinking offbaseline
      32%95% CI 27–37
      		—1.1k
      thinking on
      37%95% CI 32–42
      		clearly higher (+5 pts, 95% CI +0.4 pts to +9 pts)5.1k
      Score by effort tier · full benchmark
      Overall pass@1; vertical bars are 95% confidence intervals.
      Claude Haiku 4.5, full benchmark: pass@1 by reasoning effort tierClaude Haiku 4.5: off 32%, on 37%0%25%50%75%100%Thinking offThinking onReasoning effort tierpass@1Claude Haiku 4.5
      Show the numbers
      ModelEffort (tier)pass@195% CITasksNo answerTokens: meanmedianLatency: median per answer
      Claude Haiku 4.5thinking off · vendor default32%27–372720%1.1k6575.8 s
      Claude Haiku 4.5thinking on37%32–422720%5.1k4.6k33 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.
      Claude Haiku 4.5, full benchmark: pass@1 against mean output tokens per taskClaude Haiku 4.5: 1.1k tokens 32%, 5.1k tokens 37%0%25%50%75%100%1k2k5kMean output tokens per task (log scale)pass@1thinking offthinking onClaude Haiku 4.5

      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