Premium models are not the default lane. They are escalation tools for work where a better answer is worth more than the API bill.

The current ladder: use Sonnet 5 for daily Claude production, Opus 5 for consequential premium Claude work, and the generally available GPT-5.6 tier that fits OpenAI-native workloads. Fable 5 subscription use is now through usage credits; Mythos 5 remains approved-access only.

For the dated return analysis, read /posts/fable-5-returns-claude-sonnet-5/. For the deeper Fable-specific buying guide, read /posts/claude-fable-5-mythos-5-cost-guardrails/.

Quick Comparison

Prices are per 1 million tokens.

ModelInputOutputBatch input/outputContextBest use
Claude Fable 5$10$50$5 / $251MRestored premium escalation; not a default lane
Claude Opus 5$5$25$2.50 / $12.501MCurrent premium Claude baseline; tune effort
GPT-5.6 Sol$5$30$2.50 / $151.05MFlagship OpenAI tier
GPT-5.6 Terra$2$12$1 / $61.05MBalanced OpenAI tier
GPT-5.6 Luna$0.20$1.20$0.10 / $0.601.05MHigh-volume OpenAI tier
Claude Sonnet 5$2 intro; $3 standard$10 intro; $15 standardverify current pricing docs1MFirst Claude cost/performance test; intro ends Aug 31
GPT-5.5 standard, short context$5$30$2.50 / $15short-context laneOpenAI-native premium work
GPT-5.5 standard, long context$10$45$5 / $22.50long-context laneLong context when OpenAI fit matters
GPT-5.5 Pro, short context$30$180$15 / $90short-context Pro laneHighest-cost OpenAI Pro-tier work

Model Notes

Claude Fable 5

Anthropic suspended Fable on June 12 and says the export controls were lifted June 30. Fable returned globally on native Claude surfaces July 1. It costs 2x Opus 5, 5x Sonnet 5 at the introductory rate, and about 3.3x Sonnet 5 at the standard rate. Treat it as a high-cost guarded escalation, not a default lane.

If your account has access, re-test it only for long-horizon autonomous coding, vision-heavy reconstruction, scientific or analytical work, and expensive decisions where Opus is visibly not enough. Avoid it when zero-data-retention terms are required, when cyber/bio/chem fallback would break the workflow, or when you need predictable low-cost scale.

Claude Opus 5

Opus 5 is the practical premium Claude baseline. It costs half as much as Fable, has a 1M context window, supports five effort levels, and is the model to try before escalating to Fable.

Artificial Analysis reported Opus 5 max at $2.03 per Intelligence Index task versus Sonnet 5 max at $1.53, so Opus is not categorically cheaper. Its high/xhigh configurations can deliver stronger performance at lower task cost on specific evaluations. On AA-Briefcase, Opus 5 max cost $17.79 per task versus Fable’s $22.30; high effort cost $10.41 and still narrowly exceeded Fable, but averaged 25.7 minutes per task.

Use Opus 5 for architecture, hard debugging, code review, multi-file reasoning, and second-pass arbitration. Opus 4.8 remains available as a historical baseline.

The value comparison against GLM-5.2 remains conservative: GLM-5.2 API is about 72% lower on input and 82.4% lower on output than Opus 5 API list pricing. That makes GLM-5.2 a value lane to test, not a proven Opus replacement.

Claude Sonnet 5

Sonnet 5 is the default Claude production test. Its $2/$10 introductory API rate runs through August 31, 2026, then becomes $3/$15. The new tokenizer can produce about 30% more tokens for equivalent text, so compare complete request and agent-loop cost rather than list price alone.

Move up only when Sonnet fails in a way that matters.

GPT-5.5

GPT-5.5 standard short-context pricing is cheaper than Fable on input and output. Long-context standard pricing matches Fable input and is lower on output. GPT-5.5 Pro is much more expensive than Fable.

Compare GPT-5.5 when cached-input economics, Codex/OpenAI tooling, OpenAI-native APIs, or long-context behavior matter more than Claude’s model behavior.

GPT-5.6 family

OpenAI lists standard short-context Sol, Terra, and Luna rates at $5/$30, $2/$12, and $0.20/$1.20 per million input/output tokens. Requests above 272K input use higher long-context rates for the full request. All three have been generally available across ChatGPT, Codex, and API since July 9, with plan-dependent product access. The Luna price-cut analysis separates benchmark evidence, token subtotals, and CAR.

Use the consolidated GPT-5.6 model guide for family pricing, cache behavior, status, and benchmark provenance.

Comparable Evidence

benchmark artifact

Premium and Restricted Model Evidence

ModelProviderStatusContextInput priceOutput priceCoding signalTool-use signalBenchmark evidenceSpeedVerdictSourcesChecked
GPT-5.5OpenAIactive
Generally available prior-generation OpenAI model retained for existing integrations and comparisons.
1.05M API; 400K Codex$5.00 / 1M$30.00 / 1Mnot verifiednot verifiednot verifiednot verifiedPrimary coding seat while ChatGPT/Codex limits fit the workload.OpenAI GPT-5.5 API model page, OpenAI GPT-5.5 ChatGPT limits, Artificial Analysis: GPT-5.5, LMArena leaderboard dataset2026-06-28
GPT-5.6 SolOpenAIactive
Generally available through ChatGPT paid plans, Codex paid plans, and the OpenAI API; plan and effort options vary.
1.05M$5.00 / 1M$30.00 / 1MArtificial Analysis reports 80 on its Coding Agent Index at max effort; OpenAI reports 64.6% on SWE-bench Pro.Generally available in API and paid Codex plans; max and ultra modes are vendor-documented.
  • Artificial Analysis Intelligence Index: 59 at max effort (independent)
  • Artificial Analysis Coding Agent Index: 80 at max effort (independent)
  • SWE-bench Pro: 64.6% (vendor)
  • AIHackers repo eval: not-run (site-owned)
OpenAI announced a selected-customer Cerebras preview for July; production latency is not verified.Generally available flagship; test on real tasks and apply stronger controls for agentic or cyber work.OpenAI GPT-5.6 general availability [archive], OpenAI API pricing [archive], Artificial Analysis GPT-5.6 evaluation [archive], Agent Arena leaderboard, OpenAI GPT-5.6 availability [archive], OpenAI GPT-5.6 system card2026-08-01
Claude Sonnet 5Anthropicactive
Generally available across Claude plans, Claude Code, the Claude API, GitHub Copilot, and supported AWS paths.
1M$2.00 / 1M$10.00 / 1MAnthropic reports substantial coding and agentic gains over Sonnet 4.6; independent normalized results are pending.Available in Claude Code and the Claude API; adaptive thinking is on by default.
  • Cross-model benchmark evidence: vendor-reported; updated chart and system card preferred (vendor)
  • Artificial Analysis task cost: $1.53 per Intelligence Index task at max (independent)
  • AIHackers repo eval: not verified (site-owned)
No site-owned normalized latency result is verified.First Claude cost/performance test before Opus 5; escalate only when the premium pass changes the accepted result.Anthropic Claude Sonnet 5 launch [archive], Claude Sonnet 5 migration guide [archive], GitHub Copilot Claude Sonnet 5 launch [archive], Claude Sonnet 5 on AWS [archive], Artificial Analysis: Claude Opus 5 [archive]2026-07-25
Claude Opus 5Anthropicactive
Current premium Claude baseline; default on Max, strongest Pro model, and available through the Claude API and supported cloud platforms.
1M$5.00 / 1M$25.00 / 1MAnthropic reports major agentic-coding gains; Artificial Analysis reports joint first on its Coding Agent Index at xhigh.Thinking is on by default; five effort settings materially change cost, latency, and task performance.
  • Artificial Analysis Intelligence Index: 61 at max effort; $2.03 per task (independent)
  • AA-Briefcase: 1720 Elo / $17.79 max; 1606 Elo / $10.41 high (independent)
  • Anthropic launch evaluations: vendor-reported; configuration varies by evaluation (vendor)
  • AIHackers repo eval: not verified (site-owned)
Artificial Analysis reports high/xhigh/max AA-Briefcase runtimes of 25.7/34.3/36.2 minutes per task; Fast mode is a separate API research preview.Premium escalation for consequential agentic and knowledge work; compare complete-task cost against Sonnet 5 before default routing.Anthropic Claude Opus 5 launch [archive], What's new in Claude Opus 5 [archive], Claude Opus 5 system card [archive], Claude models overview [archive], Claude API pricing [archive], Artificial Analysis: Claude Opus 5 [archive], Artificial Analysis: Claude Opus 5 on AA-Briefcase [archive]2026-07-25
Claude Opus 4.8Anthropichistorical
Still available, but superseded by Opus 5 for current premium comparisons.
1M$5.00 / 1M$25.00 / 1MHistorical premium Claude baseline; use Opus 5 for new task-level comparisons.Still available for pinned integrations; new Claude premium routing should test Opus 5.
  • Artificial Analysis Intelligence Index v4.1: 56 (independent)
  • Artificial Analysis output speed: 57.3 tokens/s (independent)
Artificial Analysis measured 57.3 output tokens/s; provider and workload latency vary.Historical premium baseline. Use Claude Opus 5 for current Claude premium routing.Claude models overview [archive], Claude API pricing [archive], Artificial Analysis: Claude Opus 4.8 [archive], Artificial Analysis Intelligence Index v4.1, LMArena leaderboard dataset, Berkeley Function Calling Leaderboard2026-07-25
Claude Fable 5Anthropicactive
Generally available; temporary subscription allowances ended July 7 and current subscription use is through usage credits.
1M$10.00 / 1M$50.00 / 1MAnthropic reports frontier launch results; independent reproducible ranking is pending.Guarded-domain requests can refuse or fall back; verify account behavior before routing.
  • Artificial Analysis Intelligence Index: 60 at max (independent)
  • AA-Briefcase: 1574 Elo / $22.30 per task (independent)
  • AIHackers repo eval: not verified (site-owned)
Task latency varies; compare complete-task runtime before escalation.High-cost guarded escalation only; use Opus 5 as the practical Claude premium baseline.Claude models overview [archive], Claude API pricing [archive], Anthropic Fable 5 and Mythos 5 [archive], Anthropic Fable/Mythos access statement [archive], Anthropic Fable 5 redeployment [archive], Artificial Analysis: Claude Opus 5 [archive], Artificial Analysis: Claude Opus 5 on AA-Briefcase [archive]2026-07-25
Claude Mythos 5Anthropicrestricted
Restored to a set of approved US organizations; broader Glasswing access remains restricted.
1M$10.00 / 1M$50.00 / 1MGeneral coding quality is not independently verified for an accessible production route.Invitation-only research access; account and compliance approval required.
  • Independent cross-model evaluation: not verified (independent)
  • AIHackers repo eval: not verified (site-owned)
not verifiedRestricted research context, not a normal production or buying recommendation.Claude models overview [archive], Claude API pricing [archive], Anthropic Fable 5 and Mythos 5 [archive], Anthropic Fable/Mythos access statement [archive], Anthropic Fable 5 redeployment [archive]2026-07-01

Active, historical, restored, and restricted models are intentionally shown together with explicit status. Scores are not normalized across different benchmark families.

Decision Matrix

If you need…Start withEscalate to
Routine production codingSonnet 5Opus 5 if architecture or multi-file reasoning fails
Premium Claude code reviewOpus 5 highRaise effort only when measured gains justify the latency
Long-horizon autonomous codingOpus 5 highxhigh/max, then Fable only after access and guardrail tests
Vision-heavy reconstructionOpus 5GPT-5.6, Gemini, or Fable only if live output is better
OpenAI-native workflowGPT-5.6 Terra or LunaGPT-5.6 Sol when the harder tier changes the result
Cyber, bio, or chemistry workDo a guardrail test firstTrusted access or another approved model
Zero-data-retention workloadAvoid Fable/Mythos-class routesUse a model/contract that meets the requirement

Cost Sanity Check

For a 100K-input / 20K-output request:

ModelApprox cost
Claude Sonnet 5$0.40 introductory / $0.60 standard
GPT-5.6 Luna standard short context$0.044
GPT-5.6 Terra standard short context$0.44
Claude Opus 5$1.00
GPT-5.6 Sol standard short context$1.10
GPT-5.5 standard short context$1.10
GPT-5.5 standard long context$1.90
Claude Fable 5$2.00
GPT-5.5 Pro short context$6.60

If a task is routine, those deltas compound quickly. If a task is high-stakes and a better answer prevents a costly mistake, the premium can be rational.

What To Verify Before Committing Spend

  • Does Sonnet already solve the task well enough?
  • Does Opus 5 improve the answer materially, and at which effort?
  • Is Fable available through your target account, plan, cloud, and region?
  • Do its retention, usage-credit, and guardrail terms fit the workload?
  • Does the Covered Model 30-day minimum fit your data policy?
  • Does the right GPT-5.6 tier beat Claude for this workload?
  • Can you measure cost per successful task, not just benchmark score?

Sources


Last verified: August 1, 2026. Opus 5 is the current premium Claude baseline, GPT-5.6 uses separate short- and long-context rates, Fable uses credits on subscription plans, and Mythos remains restricted. Pricing, entitlements, latency, safeguards, and benchmark results can change independently.