TokenMix Research Lab · 2026-07-13
Claude Opus 5 Release Date: 4 Signals, Pricing, What's Real
Last Updated: 2026-07-13 Author: TokenMix Research Lab Data verified: 2026-07-13 - Anthropic newsroom, Claude model catalog, API pricing documentation, Opus 4.5-4.8 launch posts, Sonnet 5 launch, and Fable 5/Mythos 5 launch materials
Anthropic has not announced Claude Opus 5. No release date, price, API ID, context limit, or benchmark is confirmed.
The evidence is narrower than the rumor cycle suggests. Anthropic released Opus 4.5 on November 24, 2025, Opus 4.6 on February 5, 2026, Opus 4.7 on April 16, and Opus 4.8 on May 28. Those official dates create three intervals of 73, 70, and 42 days (Opus 4.5, Opus 4.6, Opus 4.7, Opus 4.8). But the product line changed after that: Sonnet 5 launched on June 30, while Fable 5 became Anthropic's most capable widely released model at $10/$50 per million tokens (Sonnet 5, Fable 5 and Mythos 5). The official model catalog still lists claude-opus-4-8 as the latest Opus model and contains no Opus 5 entry (Anthropic model overview).
Table of Contents
- Quick Verdict
- The Four Signals
- Release Window Scenarios
- Where Opus 5 Would Fit
- Pricing Forecast
- Cost Per Workload
- Capability and Benchmark Forecast
- API and Context Expectations
- Migration Decision
- Risks and Falsifiers
- Final Recommendation
- FAQ
- About TokenMix
- Sources
- Related Articles
Quick Verdict
The only defensible July 2026 verdict is that Opus 5 remains unannounced, while its likely product slot can be inferred more confidently than its launch date.
| Claim | Status | Evidence |
|---|---|---|
| Anthropic has officially announced Claude Opus 5 | False as of 2026-07-13 | No launch page, model-catalog row, system card, or API documentation exists |
claude-opus-5 is a valid Claude API model ID |
False as a current availability claim | Latest official Opus ID is claude-opus-4-8 |
| Opus remains an active Claude product tier | Confirmed | Anthropic describes Opus 4.8 as its current Opus-tier model |
| Anthropic has moved the Sonnet family to generation 5 | Confirmed | Sonnet 5 launched June 30, 2026 |
| Fable 5 is positioned above Opus 4.8 | Confirmed | Official catalog calls Fable 5 the next-generation option for long-running agents |
| A cadence-only release window falls in July-August 2026 | Speculation | Derived from three prior Opus intervals; not an Anthropic forecast |
| Opus 5 will keep the current $5/$25 price | Likely, not confirmed | The Opus tier has held this rate from 4.5 through 4.8 |
| Opus 5 will use a 1M-token context window | Likely, not confirmed | Current Fable, Opus, and Sonnet flagships all include 1M context |
| Published Opus 5 benchmark scores are real | False unless sourced to a future official card or reproducible evaluator | No official Opus 5 evaluation exists |
| Opus 5 could be replaced by Fable as Anthropic's premium brand | Speculation | Product architecture permits it, but Anthropic has not said this |
The Four Signals
Four signals matter, and only two point cleanly toward an eventual Opus 5 release.
| Signal | What is confirmed | What it implies | Confidence |
|---|---|---|---|
| Opus release cadence | 4.5, 4.6, 4.7, and 4.8 arrived within 185 days | Anthropic has iterated the Opus line quickly | Confirmed history |
| Sonnet generation change | Sonnet 5 replaced Sonnet 4.6 as the current Sonnet tier | A generation-5 Opus name would be consistent | Likely naming inference |
| Fable product tier | Fable 5 now sits above Opus 4.8 in capability and price | Opus 5 needs a distinct price/performance role | Confirmed fact, speculative consequence |
| Official documentation silence | Current docs list Fable 5, Opus 4.8, Sonnet 5, and Haiku 4.5 | No launch is ready to claim publicly | Confirmed as of the verification date |
The naming signal is real but incomplete. Anthropic moved Sonnet directly from 4.6 to 5, so an Opus 5 name would be unsurprising. However, Fable 5 already occupies the top public-capability position, which means Opus 5 cannot simply be described as "the strongest Claude" without colliding with Fable.
Release Window Scenarios
A cadence-only model says July-August; a product-line model says Anthropic may wait longer.
| Release | Official date | Days since prior Opus | Status |
|---|---|---|---|
| Claude Opus 4.5 | 2025-11-24 | N/A | Confirmed |
| Claude Opus 4.6 | 2026-02-05 | 73 | Confirmed, calendar calculation |
| Claude Opus 4.7 | 2026-04-16 | 70 | Confirmed, calendar calculation |
| Claude Opus 4.8 | 2026-05-28 | 42 | Confirmed, calendar calculation |
| Claude Opus 5 | Not announced | Unknown | Speculation only |
Applying the observed 42-73 day range to May 28 creates a mechanical window from July 9 through August 9. That calculation is real; treating it as a release promise would be wrong. Three intervals are a small sample, and Anthropic changed its lineup by launching Fable 5 and Sonnet 5 on June 30.
| Scenario | Earliest plausible timing | Why | Status |
|---|---|---|---|
| Cadence continues unchanged | July-August 2026 | Matches the 42-73 day historical interval | Speculation |
| Anthropic separates major launches | After the Sonnet 5 introductory period | Avoids crowding Sonnet 5 and clarifies tier economics | Speculation |
| Opus 5 becomes the middle premium tier | Q3 2026 or later | Preserves Sonnet / Opus / Fable ladder | Likely product logic, no date confirmation |
| Fable replaces new Opus releases | No Opus 5 release | Anthropic could prioritize capability-class names | Speculation |
Our base judgment is deliberately broad: late Q3 2026 is the earliest product-line window that makes sense, but confidence is low. A July or August release remains possible under cadence logic. No exact day is supportable.
Where Opus 5 Would Fit
The cleanest role for Opus 5 is between Sonnet 5 and Fable 5, not above both by default.
Anthropic's official catalog currently creates a three-step ladder: Sonnet 5 for scaled frontier work, Opus 4.8 for complex agentic and enterprise tasks, and Fable 5 for the hardest long-running agents (model overview). The Sonnet 5 review shows why a new Opus still has room: Sonnet can approach Opus 4.8 on some higher-effort tasks, while Fable costs twice as much as Opus.
| Current tier | Official positioning | Standard input/output | Likely future role | Status |
|---|---|---|---|---|
| Sonnet 5 | Speed/intelligence balance at scale | $3/$15 after 2026-08-31 | Default production model | Confirmed pricing and positioning |
| Opus 4.8 | Complex agentic coding and enterprise work | $5/$25 | Current premium middle tier | Confirmed |
| Hypothetical Opus 5 | Not published | Not published | Better premium efficiency than Fable | Speculation |
| Fable 5 | Next-generation intelligence for long-running agents | $10/$50 | Maximum public capability | Confirmed |
| Mythos 5 | Same underlying model as Fable with fewer safeguards | $10/$50 | Restricted cyber/research access | Confirmed |
This structure makes an "Opus 5 equals Fable 5" theory weak. Anthropic documents them as different product families. It could still merge or rename tiers later, but there is no evidence that it has done so.
Pricing Forecast
$5 input and $25 output per million tokens is the strongest pricing hypothesis, but it is not a published Opus 5 price.
Anthropic cut Opus pricing to $5/$25 with Opus 4.5 and kept it for 4.6, 4.7, and 4.8 (pricing documentation). The current ladder also leaves a deliberate gap between Sonnet 5 at $3/$15 and Fable 5 at $10/$50. Holding Opus at $5/$25 would preserve that ladder.
| Opus 5 price scenario | Input / 1M | Output / 1M | Evidence | Status |
|---|---|---|---|---|
| Same-tier continuity | $5 | $25 | Four consecutive Opus releases use this rate | Likely, not confirmed |
| Fable-level premium | $10 | $50 | Matches Anthropic's current highest public tier | Speculation |
| New intermediate rate | Unknown | Unknown | Anthropic could reprice the ladder | Speculation |
| Lower than Sonnet 5 | Below $3 | Below $15 | Conflicts with current tier positioning | Unlikely speculation |
| Cache hit at 10% of input | Unknown | N/A | Current Claude cache-hit multiplier is 0.1x | Likely mechanic, exact Opus 5 support unconfirmed |
| Batch at 50% off | Unknown | Unknown | Current supported Claude models receive this discount | Likely mechanic, exact Opus 5 support unconfirmed |
Do not publish $5/$25 as fact until Anthropic adds Opus 5 to its pricing table. A model name in a screenshot, gateway catalog, or client bundle would still not prove direct API pricing.
Cost Per Workload
At the same $5/$25 rate, Opus 5 would cost exactly what Opus 4.8 costs before token-efficiency changes.
The following calculations compare official current prices with a clearly labeled hypothetical Opus 5 rate. They do not claim Opus 5 pricing exists.
| Monthly workload | Sonnet 5 standard | Opus 4.8 | Hypothetical Opus 5 at $5/$25 | Fable 5 |
|---|---|---|---|---|
| 1M input + 200K output | $6 | $10 | $10 | $20 |
| 10M input + 2M output | $60 | $100 | $100 | $200 |
| 100M input + 20M output | $600 | $1,000 | $1,000 | $2,000 |
| 1B input + 200M output | $6,000 | $10,000 | $10,000 | $20,000 |
The formula for the 100M/20M workload is direct: Opus 4.8 costs 100 x $5 + 20 x $25 = $1,000; Fable 5 costs 100 x $10 + 20 x $50 = $2,000. If a future Opus 5 remains $5/$25, quality and token efficiency, not sticker price, decide whether migration saves money.
Caching changes the economics more than a small benchmark gain:
| 100M-input agent workload | Fresh input | Cache hits | Output | Monthly cost | Status |
|---|---|---|---|---|---|
| Sonnet 5: 20M fresh, 80M cached, 20M output | $60 | $24 | $300 | $384 | Confirmed current rates after Aug. 31 |
| Opus 4.8: 20M fresh, 80M cached, 20M output | $100 | $40 | $500 | $640 | Confirmed current rates |
| Opus 5 at Opus 4.8 rates | $100 | $40 | $500 | $640 | Speculation scenario |
| Fable 5: 20M fresh, 80M cached, 20M output | $200 | $80 | $1,000 | $1,280 | Confirmed current rates |
A future Opus 5 becomes economically compelling only if it reduces retries, tool steps, or output tokens enough to beat the current $640 Opus 4.8 baseline on this workload. A higher benchmark score alone does not guarantee that.
Capability and Benchmark Forecast
The responsible forecast names capability directions, not invented scores.
Opus 4.8 already targets coding agents, computer use, enterprise analysis, tool calling, and long sessions. Fable 5 extends autonomy and memory beyond that. Therefore a useful Opus 5 would need to close part of the gap while remaining cheaper or less restricted than Fable. No public evidence supports exact SWE-bench, Terminal-Bench, GPQA, OSWorld, or GDPval numbers for Opus 5.
| Capability | Minimum credible improvement | Forecast | Evidence boundary |
|---|---|---|---|
| Agentic coding | Fewer failed end-to-end tasks than Opus 4.8 | Likely product goal | No score published |
| Tool use | Fewer tool errors and redundant steps | Likely product goal | No score published |
| Computer use | Better browser/desktop reliability | Likely | Follows recent Opus direction, not confirmed |
| Long-horizon work | Longer coherent runs | Likely | Needed to sit between Opus 4.8 and Fable 5 |
| Token efficiency | Equal quality with fewer output tokens | Likely product goal | No usage multiplier published |
| Scientific reasoning | Improvement over Opus 4.8 | Speculation | Fable/Mythos may remain the research tier |
| Cyber capability | More capable with stronger safeguards | Speculation | Current Fable policy makes this sensitive |
| Benchmark sweep | Wins every major benchmark | False as a credible prediction | Frontier models have uneven strengths |
Any article publishing exact Opus 5 benchmark values today is either describing a private test that cannot be audited or inventing data. The correct placeholder is "not published."
API and Context Expectations
claude-opus-5 and a 1M-token context window are plausible conventions, not current API facts.
The official API IDs use claude-opus-4-8, claude-sonnet-5, and claude-fable-5. Current flagship models include a 1M-token context window at standard rates, while Opus 4.8 supports adaptive thinking and an effort control (models documentation). Those patterns support expectations, not availability claims.
| Field | Current verified value | Opus 5 expectation | Status |
|---|---|---|---|
| Latest Opus API ID | claude-opus-4-8 |
claude-opus-5 would be conventional |
Current Confirmed / future Speculation |
| Context window | 1M on Opus 4.8 | 1M or more | Likely, exact limit unknown |
| Max batch output | Up to 300K with beta header on Opus 4.8 | Unknown | Opus 5 unsupported claim |
| Thinking mode | Adaptive thinking on Opus 4.8 | Adaptive thinking | Likely, not confirmed |
| Effort control | Supported on Opus 4.8 | Low/medium/high or similar | Likely, not confirmed |
| Fast mode | $10/$50 on Opus 4.8 | Unknown | Speculation |
| Bedrock / Vertex / Foundry | Opus 4.8 is listed across major clouds | Rollout may lag direct API | Likely rollout pattern |
| TokenMix route | No Opus 5 route can be assumed | Verify live catalog after official launch | Confirmed operational rule |
Developers should not put claude-opus-5 into production configuration yet. A speculative model string creates avoidable 404s and can conceal fallback behavior if a gateway silently remaps unknown models.
Migration Decision
Do not delay an Opus 4.8 deployment solely because Opus 5 might appear.
The Opus 4.8 review covers a real, documented route. Waiting has a concrete opportunity cost; Opus 5 has no official date or API contract. Build migration readiness without pausing current work.
| Situation | Action now | Trigger to reconsider | Confidence |
|---|---|---|---|
| Starting a production coding agent | Benchmark Opus 4.8, Sonnet 5, and Fable 5 | Official Opus 5 API row | Confirmed recommendation |
| Already on Opus 4.8 | Keep model ID explicit and log token use | Opus 5 system card plus price | Confirmed recommendation |
| Cost-sensitive high volume | Route Sonnet first, escalate hard tasks | Opus 5 proves lower cost per success | Likely |
| Long-running autonomous research | Test Fable 5 with safeguards in mind | Opus 5 closes capability gap at lower cost | Likely |
| Compliance-sensitive deployment | Wait for retention and cloud documentation | Published data policy and regional support | Confirmed requirement |
| Gateway user | Disable silent model substitution | Live catalog exposes exact route | Confirmed requirement |
Use a 100-300 task canary when Opus 5 actually ships. Compare task success, retries, output tokens, tool errors, latency, refusal/fallback behavior, and cost per successful task. Public benchmark deltas are not a substitute for this migration test.
Risks and Falsifiers
The largest prediction risk is assuming Anthropic's naming system will remain stable after Fable and Mythos.
| Prediction | What would support it | What would falsify it | Current status |
|---|---|---|---|
| Opus 5 will exist | Official launch page, system card, or model catalog row | Anthropic retires Opus naming or designates Fable as replacement | Speculation |
| Price stays $5/$25 | Official pricing table | Any different published rate | Likely |
| Context stays at least 1M | Models API or launch docs | Published limit below 1M | Likely |
| It sits below Fable | Product comparison and lower price | Anthropic calls Opus 5 its top model above Fable | Likely |
| It launches in Q3 | Official dated release | Q3 ends without launch | Speculation |
| It improves agentic coding | Reproducible task results | No improvement or regression on production tasks | Likely goal, outcome unknown |
| It appears on all clouds immediately | Same-day cloud catalog entries | Staged or region-limited rollout | Speculation |
There is also a policy risk. Fable 5's June suspension and July redeployment showed that frontier-model availability can change after launch, and Anthropic now applies stronger safeguards and retention requirements to Mythos-class traffic (redeployment update). Opus 5 access terms cannot be inferred from model capability alone.
Final Recommendation
Treat Claude Opus 5 as a watchlist item, not a procurement assumption. The most plausible role is a $5/$25 premium tier between Sonnet 5 and Fable 5, with stronger agents and similar API mechanics to Opus 4.8. Every exact date, benchmark, model ID, and limit remains unconfirmed.
FAQ
Is Claude Opus 5 officially announced?
No. Anthropic's newsroom, model overview, pricing documentation, and system-card index contain no Claude Opus 5 entry as of July 13, 2026. The latest official Opus model is Claude Opus 4.8.
When will Claude Opus 5 be released?
There is no official date. Prior Opus intervals create a cadence-only July-August window, but Sonnet 5 and Fable 5 changed the product lineup. Late Q3 is our low-confidence product forecast, not an Anthropic announcement.
What will Claude Opus 5 cost?
Pricing has not been published. $5 input and $25 output per million tokens is the strongest hypothesis because Opus 4.5 through 4.8 share that price and it sits between Sonnet 5 and Fable 5. It must remain labeled Likely until official documentation appears.
Is Fable 5 actually Claude Opus 5?
No official source says that. Anthropic lists Fable 5 and Opus 4.8 as separate model families with different prices and product roles. Any claim that Fable is a renamed Opus 5 is speculation.
Will the API model ID be claude-opus-5?
That name is plausible but unavailable today. Anthropic uses claude-sonnet-5 and claude-fable-5, so claude-opus-5 would fit the convention. Do not deploy it until the Models API or official docs list it.
Will Opus 5 have a 1M context window?
Likely, but not confirmed. Opus 4.8, Sonnet 5, and Fable 5 currently include 1M context at standard pricing. Anthropic has not published an Opus 5 context or output limit.
Are any Claude Opus 5 benchmark scores real today?
No public score is officially verifiable today. Anthropic has not released an Opus 5 system card, and no reproducible independent evaluation can be tied to a public model ID. Treat exact numbers as unsupported.
Should I wait for Opus 5 instead of using Opus 4.8?
No for an active project. Use Opus 4.8, Sonnet 5, or Fable 5 based on current workload economics, and keep model routing configurable. Re-evaluate only after Anthropic publishes the price, API contract, and system card.
About TokenMix
TokenMix.ai provides multi-provider AI model access through one OpenAI-compatible endpoint at https://api.tokenmix.ai/v1. Check current model availability, compare pricing, and review the API documentation before selecting a production route. A future Opus 5 route should not be assumed until it appears in the live catalog.
Sources
- Anthropic Models Overview
- Anthropic API Pricing
- Anthropic Newsroom
- Introducing Claude Opus 4.5
- Introducing Claude Opus 4.6
- Introducing Claude Opus 4.7
- Introducing Claude Opus 4.8
- Introducing Claude Sonnet 5
- Claude Fable 5 and Claude Mythos 5
- Redeploying Claude Fable 5
- Claude Platform System Card Index