Skip to content

Model vs model

Claude Sonnet 5 vs Claude Haiku 4.5

Claude Sonnet 5 and Claude Haiku 4.5 are both in llmwise. What each gets on every plan, what it reads, how it's served and what happens when its provider fails, from the catalog and the code that runs them.

Model prices and specs checked against OpenRouter's Claude Sonnet 5 page, OpenRouter's Claude Haiku 4.5 page. Updated .

Short answer

Claude Haiku 4.5 gets twice as many messages: up to 250 messages a month on Pro, against up to 125 messages a month for Claude Sonnet 5. Beyond that, they read the same files and neither is easier to try. In our test runs, Claude Sonnet 5 passed 46 of the 50 prompts both answered and Claude Haiku 4.5 43; 9 prompts split them, most on coding (5 to 3).

Claude Sonnet 5 vs Claude Haiku 4.5, prompt by prompt

Every prompt Claude Sonnet 5 and Claude Haiku 4.5 both answered, compared directly, their biggest differences first. One run each, through OpenRouter: a wait depends on the provider and the load that day, so a lead under 10% counts as close.

Of the 50 prompts both answered, both passed 40, only Claude Sonnet 5 passed 6, only Claude Haiku 4.5 passed 3, and neither passed 1. Claude Sonnet 5 answered sooner on 1 of the 50 and Claude Haiku 4.5 on 45; the rest were within 10% of each other. The 50 replies cost $0.2030 on Claude Sonnet 5 and $0.0741 on Claude Haiku 4.5: 2.7× less on Claude Haiku 4.5.

The 9 prompts only one of Claude Sonnet 5 and Claude Haiku 4.5 passed

  • Evaluate an arithmetic expression, no eval (coding): Claude Sonnet 5 passed and Claude Haiku 4.5 didn't. Claude Sonnet 5: All 15 tests passed. Claude Haiku 4.5: 14 of 15 tests passed. First failure: evaluate("-2 ^ 2"): Expected values to be strictly deep-equal:

  • Parse CSV with quoted fields (coding): Claude Sonnet 5 passed and Claude Haiku 4.5 didn't. Claude Sonnet 5: All 8 tests passed. Claude Haiku 4.5: 5 of 8 tests passed. First failure: parseCsv('"line1\nline2",end\r\nnext,row\r\n'): Expected values to be strictly deep-equal:

  • Announce a second bakery shop on LinkedIn (writing): Claude Haiku 4.5 passed and Claude Sonnet 5 didn't. Claude Sonnet 5: Graded 3.5 of 5 on average (lowest 3). Claude Haiku 4.5: Graded 4.0 of 5 on average (lowest 3).

  • A product announcement with five rules (writing): Claude Sonnet 5 passed and Claude Haiku 4.5 didn't. Claude Sonnet 5: Graded 4.3 of 5 on average (lowest 4). Claude Haiku 4.5: Graded 3.7 of 5 on average (lowest 3).

  • An article in three bullets (summarization): Claude Haiku 4.5 passed and Claude Sonnet 5 didn't. Claude Sonnet 5: Graded 3.7 of 5 on average (lowest 3). Claude Haiku 4.5: Graded 4.0 of 5 on average (lowest 3).

  • An email thread in one sentence (summarization): Claude Haiku 4.5 passed and Claude Sonnet 5 didn't. Claude Sonnet 5: Graded 5.0 of 5 on average (lowest 5); but 32 words, over the 30 allowed. Claude Haiku 4.5: Graded 4.7 of 5 on average (lowest 4).

  • Average order value in August (data analysis): Claude Sonnet 5 passed and Claude Haiku 4.5 didn't. Claude Sonnet 5: Final answer 300.00: right. Claude Haiku 4.5: Final answer 275.00; expected 300.

  • A frustrated customer (customer support): Claude Sonnet 5 passed and Claude Haiku 4.5 didn't. Claude Sonnet 5: Graded 4.3 of 5 on average (lowest 4). Claude Haiku 4.5: Graded 3.3 of 5 on average (lowest 3).

  • A message with a planted instruction (customer support): Claude Sonnet 5 passed and Claude Haiku 4.5 didn't. Claude Sonnet 5: Graded 4.0 of 5 on average (lowest 3). Claude Haiku 4.5: Graded 4.3 of 5 on average (lowest 3); but 165 words, over the 150 allowed.

Job by job, the widest gaps first

  • Coding: Claude Sonnet 5 passed 5 of 5 and Claude Haiku 4.5 3 of 5. Claude Haiku 4.5 answered 1.1× sooner at the median, 2.7 s against 2.3 s. Claude Haiku 4.5 cost 3.7× less, $0.0508 against $0.0137 for the 5 replies. Their replies ran to about the same length.

  • Summarization: Claude Sonnet 5 passed 3 of 5 and Claude Haiku 4.5 5 of 5. Claude Haiku 4.5 answered 1.6× sooner at the median, 2.7 s against 1.7 s. Claude Haiku 4.5 cost 2.8× less, $0.0140 against $0.0051 for the 5 replies. Claude Sonnet 5's replies ran 47% longer, in tokens of reply, thinking not counted.

  • Customer support: Claude Sonnet 5 passed 5 of 5 and Claude Haiku 4.5 3 of 5. Claude Haiku 4.5 answered 1.3× sooner at the median, 3.5 s against 2.7 s. Claude Haiku 4.5 cost 2.5× less, $0.0182 against $0.0073 for the 5 replies. Claude Sonnet 5's replies ran 22% longer, in tokens of reply, thinking not counted.

  • Data analysis: Claude Sonnet 5 passed 4 of 5 and Claude Haiku 4.5 3 of 5. Claude Haiku 4.5 answered 1.9× sooner at the median, 5.3 s against 2.7 s. Claude Haiku 4.5 cost 2.8× less, $0.0344 against $0.0125 for the 5 replies. Claude Haiku 4.5's replies ran 150% longer, in tokens of reply, thinking not counted.

  • Writing: Claude Sonnet 5 passed 4 of 5 and Claude Haiku 4.5 4 of 5. Claude Haiku 4.5 answered 2.0× sooner at the median, 4.5 s against 2.2 s. Claude Haiku 4.5 cost 3.0× less, $0.0171 against $0.0057 for the 5 replies. Claude Sonnet 5's replies ran 65% longer, in tokens of reply, thinking not counted.

  • SQL: Claude Sonnet 5 passed 5 of 5 and Claude Haiku 4.5 5 of 5. Claude Haiku 4.5 answered 1.7× sooner at the median, 2.2 s against 1.2 s. Claude Haiku 4.5 cost 2.9× less, $0.0142 against $0.0049 for the 5 replies. Claude Sonnet 5's replies ran 35% longer, in tokens of reply, thinking not counted.

  • Agents and tool use: Claude Sonnet 5 passed 5 of 5 and Claude Haiku 4.5 5 of 5. Claude Haiku 4.5 answered 1.7× sooner at the median, 2.0 s against 1.1 s. Claude Haiku 4.5 cost 2.6× less, $0.0121 against $0.0047 for the 5 replies. Claude Haiku 4.5's replies ran 26% longer, in tokens of reply, thinking not counted.

  • RAG and answering from documents: Claude Sonnet 5 passed 5 of 5 and Claude Haiku 4.5 5 of 5. Claude Haiku 4.5 answered 1.5× sooner at the median, 1.8 s against 1.2 s. Claude Haiku 4.5 cost 2.2× less, $0.0109 against $0.0050 for the 5 replies. Claude Haiku 4.5's replies ran 56% longer, in tokens of reply, thinking not counted.

  • Math: Claude Sonnet 5 passed 5 of 5 and Claude Haiku 4.5 5 of 5. Claude Haiku 4.5 answered 1.8× sooner at the median, 3.9 s against 2.2 s. Claude Haiku 4.5 cost 2.1× less, $0.0165 against $0.0077 for the 5 replies. Claude Haiku 4.5's replies ran 71% longer, in tokens of reply, thinking not counted.

  • Translation: Claude Sonnet 5 passed 5 of 5 and Claude Haiku 4.5 5 of 5. Claude Haiku 4.5 answered 1.5× sooner at the median, 2.8 s against 1.8 s. Claude Haiku 4.5 cost 2.0× less, $0.0149 against $0.0074 for the 5 replies. Claude Haiku 4.5's replies ran 14% longer, in tokens of reply, thinking not counted.

All 50 prompts: who passed, who answered sooner, who cost less
Claude Sonnet 5 and Claude Haiku 4.5 on each prompt of our test runs
PromptResultSoonerCheaper
Turn a title into a URL slugBoth passedClaude Haiku 4.5, 1.5×took 2.3 s and 1.5 sClaude Haiku 4.5, 2.5×cost $0.0024 and $0.0010
Parse a duration like “1h 30m”Both passedClosetook 2.6 s and 2.3 sClaude Haiku 4.5, 1.7×cost $0.0036 and $0.0022
Merge overlapping intervalsBoth passedClaude Haiku 4.5, 1.4×took 2.7 s and 2.0 sClaude Haiku 4.5, 2.0×cost $0.0034 and $0.0017
Evaluate an arithmetic expression, no evalOnly Claude Sonnet 5Claude Haiku 4.5, 2.2×took 16.9 s and 7.6 sClaude Haiku 4.5, 4.0×cost $0.0226 and $0.0056
Parse CSV with quoted fieldsOnly Claude Sonnet 5Claude Haiku 4.5, 4.5×took 15.8 s and 3.5 sClaude Haiku 4.5, 5.6×cost $0.0187 and $0.0033
Announce a second bakery shop on LinkedInOnly Claude Haiku 4.5Claude Haiku 4.5, 1.8×took 4.6 s and 2.6 sClaude Haiku 4.5, 3.0×cost $0.0037 and $0.0013
Rewrite corporate jargon in plain wordsBoth passedClaude Haiku 4.5, 1.7×took 2.7 s and 1.6 sClaude Haiku 4.5, 2.7×cost $0.0022 and $0.0008
Decline a meeting and offer two timesBoth passedClaude Haiku 4.5, 1.4×took 2.8 s and 2.0 sClaude Haiku 4.5, 2.6×cost $0.0026 and $0.0010
A product announcement with five rulesOnly Claude Sonnet 5Claude Haiku 4.5, 2.0×took 4.5 s and 2.2 sClaude Haiku 4.5, 2.9×cost $0.0035 and $0.0012
Argue both sides of free busesBoth passedClaude Haiku 4.5, 1.9×took 6.6 s and 3.4 sClaude Haiku 4.5, 3.6×cost $0.0050 and $0.0014
A discount, then sales taxBoth passedClaude Haiku 4.5, 1.7×took 1.9 s and 1.1 sClaude Haiku 4.5, 2.1×cost $0.0015 and $0.0007
Pens at 3 for $4Both passedClaude Haiku 4.5, 2.9×took 6.3 s and 2.2 sClaude Haiku 4.5, 4.3×cost $0.0061 and $0.0014
Compound interest over three yearsBoth passedClosetook 3.7 s and 3.5 sClaude Haiku 4.5, 1.6×cost $0.0021 and $0.0013
Four-digit numbers whose digits sum to 9Both passedClaude Haiku 4.5, 1.6×took 4.5 s and 2.9 sClaude Haiku 4.5, 1.6×cost $0.0039 and $0.0024
The highest of three dice is a 5Both passedClaude Haiku 4.5, 1.8×took 3.9 s and 2.2 sClaude Haiku 4.5, 1.6×cost $0.0028 and $0.0018
An article in three bulletsOnly Claude Haiku 4.5Claude Haiku 4.5, 1.5×took 2.6 s and 1.8 sClaude Haiku 4.5, 2.5×cost $0.0027 and $0.0011
An email thread in one sentenceOnly Claude Haiku 4.5Claude Haiku 4.5, 1.8×took 2.3 s and 1.3 sClaude Haiku 4.5, 2.6×cost $0.0019 and $0.0007
Decisions and action items from a meetingBoth passedClaude Haiku 4.5, 1.7×took 2.8 s and 1.7 sClaude Haiku 4.5, 3.1×cost $0.0033 and $0.0011
A quarterly memo for the CEOBoth passedClaude Haiku 4.5, 1.9×took 3.0 s and 1.5 sClaude Haiku 4.5, 2.7×cost $0.0033 and $0.0012
A study with a negative resultBoth passedClaude Sonnet 5, 1.2×took 2.7 s and 3.4 sClaude Haiku 4.5, 2.8×cost $0.0028 and $0.0010
The region with the most revenueBoth passedClaude Haiku 4.5, 2.4×took 6.4 s and 2.7 sClaude Haiku 4.5, 1.5×cost $0.0041 and $0.0026
Average order value in AugustOnly Claude Sonnet 5Claude Haiku 4.5, 1.9×took 3.9 s and 2.1 sClaude Haiku 4.5, 1.8×cost $0.0041 and $0.0023
Revenue change from July to AugustBoth passedClaude Haiku 4.5, 1.3×took 5.3 s and 4.2 sClaude Haiku 4.5, 2.8×cost $0.0062 and $0.0022
A median, filtered two waysBoth passedClaude Haiku 4.5, 2.5×took 4.9 s and 2.0 sClaude Haiku 4.5, 2.2×cost $0.0036 and $0.0016
Correlation between ad spend and sign-upsNeither passedClaude Haiku 4.5, 2.8×took 13.8 s and 5.0 sClaude Haiku 4.5, 4.5×cost $0.0165 and $0.0037
A late orderBoth passedClaude Haiku 4.5, 1.7×took 3.5 s and 2.0 sClaude Haiku 4.5, 2.6×cost $0.0036 and $0.0014
A return inside the windowBoth passedClaude Haiku 4.5, 1.2×took 2.9 s and 2.4 sClaude Haiku 4.5, 2.4×cost $0.0032 and $0.0013
A frustrated customerOnly Claude Sonnet 5Claude Haiku 4.5, 1.5×took 4.4 s and 3.0 sClaude Haiku 4.5, 2.6×cost $0.0041 and $0.0016
A refund request outside the windowBoth passedClaude Haiku 4.5, 1.2×took 3.2 s and 2.7 sClaude Haiku 4.5, 2.4×cost $0.0031 and $0.0013
A message with a planted instructionOnly Claude Sonnet 5Claude Haiku 4.5, 1.4×took 4.6 s and 3.2 sClaude Haiku 4.5, 2.3×cost $0.0041 and $0.0018
A delivery message into SpanishBoth passedClaude Haiku 4.5, 1.8×took 2.7 s and 1.5 sClaude Haiku 4.5, 2.6×cost $0.0029 and $0.0011
A product description into FrenchBoth passedClaude Haiku 4.5, 1.8×took 2.5 s and 1.4 sClaude Haiku 4.5, 2.5×cost $0.0029 and $0.0011
A meeting note into GermanBoth passedClaude Haiku 4.5, 1.5×took 2.8 s and 1.8 sClaude Haiku 4.5, 2.6×cost $0.0027 and $0.0010
Idioms into natural JapaneseBoth passedClaude Haiku 4.5, 1.2×took 4.1 s and 3.3 sClaude Haiku 4.5, 1.5×cost $0.0028 and $0.0019
A lease clause into Brazilian PortugueseBoth passedClosetook 3.8 s and 3.5 sClaude Haiku 4.5, 1.7×cost $0.0036 and $0.0022
Customers in one countryBoth passedClaude Haiku 4.5, 1.6×took 1.8 s and 1.1 sClaude Haiku 4.5, 3.1×cost $0.0020 and $0.0007
Count orders by statusBoth passedClaude Haiku 4.5, 2.4×took 2.0 s and 0.8 sClaude Haiku 4.5, 3.1×cost $0.0020 and $0.0006
Revenue by categoryBoth passedClaude Haiku 4.5, 2.8×took 3.5 s and 1.2 sClaude Haiku 4.5, 2.9×cost $0.0031 and $0.0011
Every customer, even those without ordersBoth passedClosetook 2.2 s and 2.1 sClaude Haiku 4.5, 3.0×cost $0.0030 and $0.0010
Monthly revenue with a running totalBoth passedClaude Haiku 4.5, 1.9×took 3.0 s and 1.6 sClaude Haiku 4.5, 2.7×cost $0.0042 and $0.0016
A fact from one sectionBoth passedClaude Haiku 4.5, 1.6×took 1.7 s and 1.1 sClaude Haiku 4.5, 2.1×cost $0.0019 and $0.0009
Core hours and start timesBoth passedClaude Haiku 4.5, 1.8×took 2.1 s and 1.1 sClaude Haiku 4.5, 2.4×cost $0.0022 and $0.0009
Two sections in one answerBoth passedClaude Haiku 4.5, 1.4×took 1.8 s and 1.3 sClaude Haiku 4.5, 2.1×cost $0.0022 and $0.0011
A later amendment changes the answerBoth passedClaude Haiku 4.5, 1.1×took 2.1 s and 1.9 sClaude Haiku 4.5, 2.1×cost $0.0026 and $0.0012
A question the handbook doesn't answerBoth passedClaude Haiku 4.5, 1.5×took 1.8 s and 1.2 sClaude Haiku 4.5, 2.1×cost $0.0019 and $0.0009
Pick the tool and work out the dateBoth passedClaude Haiku 4.5, 1.6×took 1.7 s and 1.1 sClaude Haiku 4.5, 2.1×cost $0.0018 and $0.0008
Convert a currencyBoth passedClaude Haiku 4.5, 2.0×took 2.0 s and 1.0 sClaude Haiku 4.5, 2.1×cost $0.0018 and $0.0008
Book a meeting from a sentenceBoth passedClaude Haiku 4.5, 4.7×took 5.7 s and 1.2 sClaude Haiku 4.5, 3.6×cost $0.0037 and $0.0010
Search, but don't bookBoth passedClaude Haiku 4.5, 1.7×took 1.9 s and 1.1 sClaude Haiku 4.5, 2.1×cost $0.0020 and $0.0010
Two calls with a unit conversionBoth passedClaude Haiku 4.5, 3.1×took 3.8 s and 1.2 sClaude Haiku 4.5, 2.7×cost $0.0028 and $0.0011

Claude Sonnet 5 vs Claude Haiku 4.5 in our test runs

Claude Sonnet 5 and Claude Haiku 4.5 on the same prompts, job by job: how many replies passed their check.

Based on 100 of our test runs on , through OpenRouter with the app's own prompt and settings. Updated .

Claude Sonnet 5 and Claude Haiku 4.5 in our test runs, job by job
JobClaude Sonnet 5Claude Haiku 4.5
Coding5 of 53 of 5
Writing4 of 54 of 5
Math5 of 55 of 5
Summarization3 of 55 of 5
Data analysis4 of 53 of 5
Customer support5 of 53 of 5
Translation5 of 55 of 5
SQL5 of 55 of 5
RAG and answering from documents5 of 55 of 5
Agents and tool use5 of 55 of 5
All jobs46 of 50 · 1 message each · $0.0041 a reply43 of 50 · 1 message each · $0.0015 a reply

The same prompts, sent to both through llmwise's own pipeline and checked the same way: five for each job. A bold count is the one that passed more. Messages: what each reply counted as on Pro. Cost: what OpenRouter charged us per reply, on average. Every prompt, and how it's scored.

Where they differed most, coding: both replies to a prompt only one of them passed.

A hard prompt: “Evaluate an arithmetic expression, no eval”, in full.

  • Claude Sonnet 5

    Passed: All 15 tests passed.

    ```javascript
    function evaluate(expression) {
      // ---- Tokenizer ----
      const tokens = [];
      let i = 0;
      const s = expression;
      while (i < s.length) {
        const c = s[i];
        if (c === ' ' || c === '\t' || c === '\n' || c === '\r') {
          i++;
          continue;
        }
        if ('+-*/^()'.includes(c)) {
          tokens.push(c);
          i++;
          continue;
        }
        if ((c >= '0' && c <= '9') || c === '.') {
          let start = i;
          let seenDot = false;
          let seenDigit = false;
          while (i < s.length && ((s[i] >= '0' && s[i] <= '9') || s[i] === '.')) {
            if (s[i] === '.') {
              if (seenDot) break;
              seenDot = true;
            } else {
              seenDigit = true;
            }
            i++;
          }
          const raw = s.slice(start, i);
          if (!seenDigit || !/^\d*\.?\d*$/.test(raw)) {
            throw new Error(`Invalid number: "${raw}"`);
          }
          tokens.push({ type: 'num', value: parseFloat(raw) });
          continue;
        }
        throw new Error(`Unexpected character "${c}" at position ${i}`);
      }
    
      // ---- Parser ----
      let pos = 0;
    
      function peek() {
        return tokens[pos];
      }
    
      function next() {
        return tokens[pos++];
      }
    
      function expect(tok) {
        const t = next();
        if (t !== tok) {
          throw new Error(`Expected "${tok}" but got "${t === undefined ? 'end of input' : (typeof t === 'object' ? t.value : t)}"`);
        }
      }
    
      function parseExpr() {
        let value = parseTerm();
        while (peek() === '+' || peek() === '-') {
          const op = next();
          const rhs = parseTerm();
          value = op === '+' ? value + rhs : value - rhs;
        }
        return value;
      }
    
      function parseTerm() {
        let value = parseUnary();
        while (peek() === '*' || peek() === '/') {
          const op = next();
          const rhs = parseUnary();
          if (op === '*') {
            value = value * rhs;
          } else {
            value = value / rhs;
          }
        }
        return value;
      }
    
      function parseUnary() {
        if (peek() === '-') {
          next();
          return -parseUnary();
        }
        if (peek() === '+') {
          next();
          return parseUnary();
        }
        return parsePower();
      }
    
      function parsePower() {
        const base = parseAtom();
        if (peek() === '^') {
          next();
          const exponent = parseUnary(); // right-associative, allows unary minus
          return Math.pow(base, exponent);
        }
        return base;
      }
    
      function parseAtom() {
        const t = peek();
        if (t === undefined) {
          throw new Error('Unexpected end of input');
        }
        if (t === '(') {
          next();
          const value = parseExpr();
          expect(')');
          return value;
        }
        if (typeof t === 'object' && t.type === 'num') {
          next();
          return t.value;
        }
        throw new Error(`Unexpected token "${typeof t === 'object' ? t.value : t}"`);
      }
    
      if (tokens.length === 0) {
        throw new Error('Empty expression');
      }
    
      const result = parseExpr();
    
      if (pos !== tokens.length) {
        const remaining = tokens[pos];
        throw new Error(`Unexpected token "${typeof remaining === 'object' ? remaining.value : remaining}" after expression`);
      }
    
      return result;
    }
    ```

    760 tokens in, 2,110 out (834 of them reasoning) · 16.9 s · $0.0226 · 1 message on Pro · answered by anthropic/claude-sonnet-5 via Claude Platform on AWS ·

  • Claude Haiku 4.5

    Failed: 14 of 15 tests passed. First failure: evaluate("-2 ^ 2"): Expected values to be strictly deep-equal:

    ```javascript
    function evaluate(expression) {
      const tokens = tokenize(expression);
      const parser = new Parser(tokens);
      const result = parser.parseExpression();
      
      if (parser.pos !== tokens.length) {
        throw new Error("Unexpected tokens after expression");
      }
      
      return result;
    }
    
    function tokenize(expression) {
      const tokens = [];
      let i = 0;
      
      while (i < expression.length) {
        const ch = expression[i];
        
        if (/\s/.test(ch)) {
          i++;
          continue;
        }
        
        if (/\d/.test(ch)) {
          let num = '';
          while (i < expression.length && /[\d.]/.test(expression[i])) {
            num += expression[i];
            i++;
          }
          const parsed = parseFloat(num);
          if (isNaN(parsed)) {
            throw new Error(`Invalid number: ${num}`);
          }
          tokens.push({ type: 'NUMBER', value: parsed });
          continue;
        }
        
        if ('+-*/^()'.includes(ch)) {
          tokens.push({ type: ch, value: ch });
          i++;
          continue;
        }
        
        throw new Error(`Unexpected character: ${ch}`);
      }
      
      return tokens;
    }
    
    class Parser {
      constructor(tokens) {
        this.tokens = tokens;
        this.pos = 0;
      }
      
      peek() {
        return this.tokens[this.pos];
      }
      
      consume() {
        return this.tokens[this.pos++];
      }
      
      parseExpression() {
        return this.parseAddSub();
      }
      
      parseAddSub() {
        let left = this.parseMulDiv();
        
        while (this.peek() && (this.peek().type === '+' || this.peek().type === '-')) {
          const op = this.consume().type;
          const right = this.parseMulDiv();
          left = op === '+' ? left + right : left - right;
        }
        
        return left;
      }
      
      parseMulDiv() {
        let left = this.parsePower();
        
        while (this.peek() && (this.peek().type === '*' || this.peek().type === '/')) {
          const op = this.consume().type;
          const right = this.parsePower();
          left = op === '*' ? left * right : left / right;
        }
        
        return left;
      }
      
      parsePower() {
        let left = this.parseUnary();
        
        if (this.peek() && this.peek().type === '^') {
          this.consume();
          const right = this.parsePower();
          left = Math.pow(left, right);
        }
        
        return left;
      }
      
      parseUnary() {
        if (this.peek() && this.peek().type === '-') {
          this.consume();
          return -this.parseUnary();
        }
        
        if (this.peek() && this.peek().type === '+') {
          this.consume();
          return this.parseUnary();
        }
        
        return this.parsePrimary();
      }
      
      parsePrimary() {
        const token = this.peek();
        
        if (!token) {
          throw new Error("Unexpected end of expression");
        }
        
        if (token.type === 'NUMBER') {
          this.consume();
          return token.value;
        }
        
        if (token.type === '(') {
          this.consume();
          const result = this.parseExpression();
          
          if (!this.peek() || this.peek().type !== ')') {
            throw new Error("Missing closing parenthesis");
          }
          
          this.consume();
          return result;
        }
        
        throw new Error(`Unexpected token: ${token.value}`);
      }
    }
    ```

    599 tokens in, 1,007 out · 7.6 s · $0.0056 · 1 message on Pro · answered by anthropic/claude-haiku-4.5 via Amazon Bedrock ·

Claude Sonnet 5 and Claude Haiku 4.5 on every plan

Whether the one-time free trial reaches each model, then each paid plan's messages on it.

Messages on Claude Sonnet 5 and Claude Haiku 4.5, plan by plan
PlanPriceClaude Sonnet 5Claude Haiku 4.5
Free$0In the one-time trial of 5 messagesIn the one-time trial of 5 messages
Pro$20 a monthUp to 125 a monthUp to 250 a month
Max$50 a monthUp to 400 a monthUp to 800 a month
Ultra$100 a monthUp to 900 a monthUp to 1,800 a month
Studio$200 a monthUp to 2,000 a monthUp to 4,000 a month

Prices don't include tax, which is added where it applies and shown before you pay. A paid plan's month is one allowance shared by every model, so each monthly count is the most you get if all of it goes to that model. It renews each billing period; everyday models refill daily at 00:00 UTC. Long chats count more per reply. How pricing works.

Every limit is published. Paid plans also have a monthly fair-use limit on AI cost: Pro $7.50, Max $20, Ultra $42, Studio $85. Using every message on your plan at typical sizes stays under it; very large messages and heavy research use it faster. Every limit, explained.

What differs

  • Messages on Pro

    Claude Haiku 4.5 gets twice as many messages: up to 250 messages a month on Pro, against up to 125 messages a month for Claude Sonnet 5.

  • Context window

    Claude Sonnet 5 takes up to 1M tokens; Claude Haiku 4.5 up to 200K tokens. A chat in llmwise holds up to 200k tokens, which fits in either, so the difference shows only through each maker's own API.

  • Images and PDFs

    Both read images. Both take a PDF as the whole file, pages and all.

  • On Free

    Both are in the free trial.

  • Where messages go

    Both are Claude models, served the same way. Sent to Anthropic directly.

Fact by fact

Claude Sonnet 5 and Claude Haiku 4.5, fact by fact
FactClaude Sonnet 5Claude Haiku 4.5
Context window1M tokens200K tokens
Reads imagesYesYes
PDFsWhole fileWhole file
ReasoningYesNo
API price (September 2026)$2.00 in / $10.00 out per million tokens$1.00 in / $5.00 out per million tokens
A typical message at API prices (4,000 tokens in, 700 out)$0.0150$0.0075
A $10 top-up adds100 messages200 messages
Where a message goesSent to Anthropic directly.Sent to Anthropic directly.
If the provider failsIf Anthropic fails before the reply starts (an overload, a server error, a dropped connection), llmwise sends the same request to Claude Sonnet 5 through OpenRouter instead.If Anthropic fails before the reply starts (an overload, a server error, a dropped connection), llmwise sends the same request to Claude Haiku 4.5 through OpenRouter instead.
Anthropic's safety fallbackDoesn't applyDoesn't apply
API prices are what our model catalog lists (Claude: Anthropic's list price). In llmwise you pay per message, not per token: the counts above are what you get.

Claude Sonnet 5 or Claude Haiku 4.5?

From the facts above and our test runs: the rest is how their answers suit your work, which one chat can show you.

  • Pick Claude Haiku 4.5: more messages for the money, up to 250 messages a month on Pro.

Each model's page, the families, and other pairs

Claude Sonnet 5 vs Claude Haiku 4.5 is one pair of models. The page below covers the whole family.

Questions

Is Claude Sonnet 5 or Claude Haiku 4.5 cheaper in llmwise?

Claude Haiku 4.5 gets twice as many messages: up to 250 messages a month on Pro, against up to 125 messages a month for Claude Sonnet 5. Every paid plan's monthly allowance is shared by all models, so each count is the most you get if it all goes to that model.

Can I try Claude Sonnet 5 and Claude Haiku 4.5 for free?

Yes: both are in the free trial of 5 messages.

Which has the bigger context window, Claude Sonnet 5 or Claude Haiku 4.5?

Claude Sonnet 5: 1M tokens, against 200K tokens. A chat in llmwise holds up to 200k tokens, which fits in either, so the difference shows only through each maker's own API.

Can I use Claude Sonnet 5 and Claude Haiku 4.5 in the same chat?

Yes. Pick Claude Sonnet 5 for one message and Claude Haiku 4.5 for the next; the second sees the whole chat, including the first one's answer.

Which did better in your test runs, Claude Sonnet 5 or Claude Haiku 4.5?

On the same 50 prompts, run on September 27, 2026, Claude Sonnet 5 passed 46 and Claude Haiku 4.5 passed 43. The table on this page has each job, and every reply is published.

Claude, GPT, Gemini, DeepSeek, Grok, Kimi, and GLM, in one chat.

See what a message costs before you send it. Free is 5 messages to try; sign in with an email link, no password or card.