Cost comparison · As of 2026-11-19

GPT-6.2 Sol vs Claude Fable 5.2 for Coding Cost: Which Is Cheaper?

As of November 19, 2026, official list prices put gpt-6.2-sol and claude-fable-5-2 at exactly 4x apart on both input and output, and a coding session with 2,000,000 input tokens (1,500,000 of them cache hits) and 200,000 output tokens bills about $3.15 on Sol versus $12.30 on Fable 5.2. Fable 5.2 still leads the Artificial Analysis Intelligence Index while Sol scores 1,259 against 1,271 on AA-3D, so the usual split is Sol for daily coding and Fable for the tasks you are stuck on.

Updated 2026-11-19

Switch models on the same endpoint and measure the cost per token yourself; check /models for callable models.

#4x input and output#2x cache reads#50% batch discount#Sol first, Fable on hard bugs

Four numbers you can put straight into a budget

4x

Input and output price gap

The official OpenAI API list price for gpt-6.2-sol is $2.00 input and $10.00 output; the official Anthropic list price for claude-fable-5-2 is $8 input and $40 output, exactly 4x on both.

2x

Cache read price gap

Cache reads are $0.10 on Sol and $0.20 on Fable 5.2 at official list prices. Fable 5.2 also lists $10 for a 5-minute cache write and $16 for a 1-hour write; Sol lists $2.50 for 30 minutes of retention and $5 for 12 hours.

$3.15/$12.30

2M input plus 200K output

Worked example: 2,000,000 input tokens (summed across the session's requests, each under 272K input tokens, with 1,500,000 cache hits) plus 200,000 output tokens costs $3.15 on Sol and $12.30 on Fable 5.2 at the official Standard short-context rates.

1259 / 1271

AA-3D leaderboard gap

On Artificial Analysis AA-3D, gpt-6.2-sol scores 1,259 and claude-fable-5-2 scores 1,271, while gpt-6.2-astra leads the same board at 1,318.

What one coding task bills on each side

As of November 19, 2026, both gpt-6.2-sol and claude-fable-5-2 publish every line of their pricing, yet almost nobody puts the two into the same task and adds it up. gpt-6.2-sol shipped with GPT-6.2 on 2026-11-17 at an official OpenAI list price of $2.00 input, $0.10 cached input and $10.00 output; claude-fable-5-2 shipped on 2026-11-09 at an official Anthropic list price of $8 input, $40 output and $0.20 cache read.

People searching GPT-6.2 Sol vs Claude Fable 5.2 for coding cost want a conversion they can recompute: what the same session bills on each side, and what the extra money buys. Three things settle it: the official rates and multipliers, one worked calculation you can refill with your own token counts, and a routing recommendation.

Four November 2026 changes: new models, batch discounts, two speed tiers

On 2026-11-09 Anthropic released claude-fable-5-2: 1M context, 128K max output, Adaptive thinking (always on), default effort high, Jun 2026 knowledge cutoff, official pricing $8 input and $40 output, plus a 50% Batch API discount on input and output. On 2026-11-17 OpenAI released gpt-6.2-sol: 1,050,000 context, 922,000 max input, 128,000 max output, Jun 01, 2026 knowledge cutoff, official pricing $2 input, $0.10 cached input and $10 output, with Batch at 50% of Standard and Fast at twice Standard.

Artificial Analysis published the same day: Sol gained 2 points to 57 on the Intelligence Index but lost 3 points on the Coding Agent Index against GPT-6.1 Sol, driven by a 14-point drop on SWE-Atlas-QnA-AA. Sol's Ultrafast tier costs 3 times Standard through February 28, 2027 and 6 times Standard from March 1, 2027 (Pacific time), so never price it as the default tier.

Three November 2026 milestones: Fable 5.2, Sol, and the 2026-11-19 check state

2026-11-09

Anthropic releases claude-fable-5-2: 1M context, 128K max output, Adaptive thinking (always on), default effort high, official pricing $8 input and $40 output with $0.20 cache reads, plus a 50% Batch API discount on input and output.

2026-11-17

OpenAI releases gpt-6.2-sol: 1,050,000 context, 922,000 max input, 128,000 max output, official pricing $2.00 input, $0.10 cached input and $10.00 output; a single request above 272K input tokens pays 2x input and cache rates and 1.5x output for the whole request.

2026-11-19

What checks out on 2026-11-19: both models are callable on QCode, gpt-6.2-sol through https://api.qcode.cc/openai and claude-fable-5-2 through https://api.qcode.cc/api, with the endpoints unchanged and live rates on qcode.cc/models; the same day also leaves one pricing milestone ahead, as Sol's Ultrafast tier stays at 3x Standard through February 28, 2027 and becomes 6x Standard from March 1, 2027 (Pacific time).

Confirmed numbers vs unpublished or community-only claims

Confirmed (vendor documentation and independent benchmark text)

  1. 1Rates per 1M tokens at Standard short context: the official OpenAI API list price for gpt-6.2-sol is $2.00 input, $0.10 cached input, $10.00 output, $2.50 for a 30-minute cache write and $5 for 12 hours; the official Anthropic list price for claude-fable-5-2 is $8 input, $40 output, $10 for a 5-minute cache write, $16 for a 1-hour cache write and $0.20 cache read, with a 50% Batch API discount on input and output.
  2. 2Surcharge rules: for Sol, a single request above 272K input tokens pays 2x on input and cache rates and 1.5x on output for the whole request; Fast is 2x Standard; Ultrafast is 3x Standard through February 28, 2027 and 6x Standard from March 1, 2027 (Pacific time).
  3. 3Specs: Sol has 1,050,000 context, 128,000 max output, a Jun 01, 2026 knowledge cutoff, and the API rejects requests that set reasoning effort to none; Fable 5.2 has 1M context, 128K max output, Adaptive thinking (always on), default effort high, a Jun 2026 knowledge cutoff and no retirement sooner than 2027-11-09.
  4. 4Independent evaluation (Artificial Analysis, 2026-11-17): AA-3D gives gpt-6.2-astra 1,318, claude-fable-5-2 1,271 and gpt-6.2-sol 1,259; AA-Robotics v0.1 gives gpt-6.2-astra 79, gpt-6.2-sol 78 and claude-fable-5-2 75.

Unpublished details and community claims

  1. 1No official SWE-bench score has been published for either model, so neither should be compared on SWE-bench; Artificial Analysis runs its Coding Agent Index in a Codex harness at max effort, which is not the same as your own repository.
  2. 2A price gap is not a capability gap, in either direction: the only checkable numbers are the leaderboards above, where Fable 5.2 still leads the Intelligence Index and Sol trails by 12 points on AA-3D.
  3. 3Anthropic's Fable 5.2 pricing page lists input, output, two cache write tiers, cache read and the Batch discount, with no long-context surcharge clause, while OpenAI publishes the 272K rule for Sol; the two use different pricing structures.
  4. 4The tokenizers differ, so the same code does not tokenize to the same count; complaints that Fable weekly caps burn fast (for example Fable5.2 is unusable for pro plan.) describe subscription limits, not API list prices.

Two ways to spend: all-Fable 5.2 vs Sol by default, Fable when stuck

Run the whole session on claude-fable-5-2

Keeping every request on Fable 5.2 suits complex architecture, hard debugging and work that needs the highest single-pass success rate, but the same token count costs 4x the official input and output price, or $12.30 in the worked example. Batching the requests through the Batch API applies the official 50% discount and leaves $4 input and $20 output, still above Sol's Standard rate.

Default to gpt-6.2-sol, switch to Fable 5.2 when stuck

Daily coding, completion, refactors and test generation run on Sol at the official $2.00 input and $10.00 output list price, and only the part that is genuinely stuck moves to Fable 5.2. The same official-rate example bills $3.15 on Sol, roughly a quarter of all-Fable, while the two sit 12 points apart on AA-3D.

Recompute the bill in three steps, then wire both models into your CLI

  1. 1Fix the basis first: Standard tier, short context (input at or below 272K), official list rates per 1M tokens, with Sol at $2.00 input, $0.10 cached input and $10.00 output and Fable 5.2 at $8 input, $0.20 cache read and $40 output.
  2. 2Drop your own token counts into those official rates: for a session with 2,000,000 input tokens (summed across the session's requests, each under 272K input tokens, 1,500,000 of them cache hits) and 200,000 output tokens, Sol is 0.5M×$2 + 1.5M×$0.10 + 0.2M×$10 = $3.15 and Fable 5.2 is 0.5M×$8 + 1.5M×$0.20 + 0.2M×$40 = $12.30; this is an illustration at official list prices, excludes cache write fees, and real token counts vary by task.
  3. 3Point your tools at both: in ~/.codex/config.toml set wire_api = "responses", base_url = "https://api.qcode.cc/openai" and model = "gpt-6.2-sol"; in Claude Code set ANTHROPIC_BASE_URL=https://api.qcode.cc/api and switch with /model claude-fable-5-2; both use the same key, so run the same task on each and read the bill afterwards.

Run both models on one QCode key

QCode serves gpt-6.2-sol and claude-fable-5-2. The GPT family uses the OpenAI-compatible endpoints: Codex CLI sets base_url = "https://api.qcode.cc/openai" with wire_api = "responses" and no /v1 suffix, while the OpenAI SDK, Cline, OpenCode and WorkBuddy use https://api.qcode.cc/openai/v1; the Claude family uses the Anthropic Messages endpoint, so Claude Code sets ANTHROPIC_BASE_URL=https://api.qcode.cc/api.

Billing is per token, with live per-model rates on /models. Not ready for a monthly plan? Pay-as-you-go Packs start at $4 for a one-time quota you use within 7 days; if you already have a Personal Plan, add a booster pack when you need more quota.

Frequently asked questions

What does one coding task actually cost on gpt-6.2-sol versus claude-fable-5-2?

At official list prices, the worked session with 2,000,000 input tokens (summed across the session's requests, each under 272K input tokens, with 1,500,000 cache hits) and 200,000 output tokens bills $3.15 on Sol and $12.30 on Fable 5.2, about 4x. The two expressions are 0.5M×$2 + 1.5M×$0.10 + 0.2M×$10 against 0.5M×$8 + 1.5M×$0.20 + 0.2M×$40, so you can swap in your own input, cache-hit and output counts.

Why is the gap 4x on input and output but only 2x on cache reads?

Both vendors price cache reads far below fresh input: Sol charges $0.10 for cache reads against $2.00 input, and Fable 5.2 charges $0.20 against $8 input, so the cache-read line itself is only a 2x gap. The more tokens hit cache, the more of the bill lands on that 2x line, so the total ratio moves from 4x toward 2x rather than staying at 4x. With zero cache hits (2,000,000 input tokens plus 200,000 output tokens), Sol costs $6.00 and Fable 5.2 costs $24.00, exactly 4x; the worked example above (1,500,000 input tokens cached) is $3.15 against $12.30, about 3.9x; and if every input token hits cache, the input line alone is exactly 2x.

Does batching the requests close the gap?

No. OpenAI's official Batch tier is 50% of Standard, which puts Sol at $1.00 input, $5.00 output and $0.05 cached input; Anthropic's official Batch API applies the same 50% discount to input and output, putting Fable 5.2 at $4 input and $20 output. Both sides are discounted by the same proportion, so the 4x ratio holds and only the absolute amount halves.

What happens when a single request goes above 272K input tokens?

On Sol, a request above 272K input tokens pays 2x the short-context input and cache rates and 1.5x the output rate for the entire request, not just the part past the threshold, and max input is 922,000 tokens. Fable 5.2's pricing page lists input, output, two cache write tiers, cache read and the Batch discount with no comparable long-context clause, but its context is also 1M, so long sessions still deserve a token-count check.

How far apart are the capabilities, and is 4x worth paying?

It depends on the work. On AA-3D, Sol scores 1,259 and Fable 5.2 scores 1,271, a 12-point gap; on AA-Robotics v0.1, Sol scores 78 and Fable 5.2 scores 75; yet Fable 5.2 still leads the Intelligence Index while Sol sits at 57. The same Artificial Analysis article notes that Sol lost 3 points on the Coding Agent Index against GPT-6.1 Sol, driven by a 14-point drop on SWE-Atlas-QnA-AA, and its hallucination figures cover only the two OpenAI models: Sol sits at 55% (gpt-6.2-astra is 59%, both AA-Omniscience hallucination rates at max effort). The practical rule is daily coding on Sol, and paying the 4x for Fable on complex architecture and hard debugging.

How do I switch between the two models with one QCode key?

By protocol. gpt-6.2-sol uses the OpenAI-compatible endpoints: base_url = "https://api.qcode.cc/openai" in Codex CLI with wire_api = "responses", and https://api.qcode.cc/openai/v1 in the OpenAI SDK or OpenCode. claude-fable-5-2 uses the Anthropic Messages endpoint: set ANTHROPIC_BASE_URL=https://api.qcode.cc/api in Claude Code and type /model claude-fable-5-2. One key covers both, with live rates on /models.

Sources

OpenAI: API pricing page (developers.openai.com/api/docs/pricing, fetched 2026-11-19) and the gpt-6.2-sol model page (fetched 2026-11-19); Anthropic: Claude Fable 5.2 overview and pricing page (platform.claude.com/docs/en/models/fable-5-2/overview, fetched 2026-11-19); Artificial Analysis: the 2026-11-17 GPT-6.2 Astra and Sol article; community: user posts on X, Reddit and Zenn between 2026-11-14 and 2026-11-19, cited as community claims only.

Do the math first, then run both models on one key

Create a QCode key to call gpt-6.2-sol and claude-fable-5-2 per token, with live rates on /models; if the project budget is tight, start with a 7-day pay-as-you-go pack.

Not sure where to start?

Try it for a week

Small Pack: $4 for $30 of quota, valid 7 days

Use it every day

Starter: $9.00/month with $5 of quota every day — just $0.060 per $1 of quota, vs about $0.113 for the Medium Pack