What Is Claude Opus 5? Anthropic’s Next Flagship, Codenamed Honeycomb

Claude Opus 5 is Anthropic's unreleased next-generation Opus flagship, first surfaced through a research model called "Honeycomb EAP" that briefly appeared inside the Cursor model picker around July 9, 2026 and was pulled within hours. As of July 15, 2026, there is no official Anthropic announcement, no claude-opus-5 API model ID in Anthropic's documentation, and no published system card. The current Opus flagship on the API remains Claude Opus 4.8, released May 28, 2026 at $5 per million input tokens and $25 per million output tokens. Community leaks aggregated from Cursor sightings, a reported Google Vertex AI listing, and multiple X posts point to a launch window between mid-July and early August 2026, with a 1 million token context window and an "xhigh" reasoning effort mode.

Key Takeaways

• Claude Opus 5 has not been officially announced by Anthropic. The current Opus flagship is Claude Opus 4.8.
• A model called "Claude Honeycomb EAP" briefly appeared in Cursor on July 9, 2026 and is widely believed to be an early Opus 5 preview, per a detailed leak thread from @LuminaXspace.
• Reported specs from leaks: 1 million token context window, an "xhigh" reasoning effort setting, and a safety fallback that routes to Claude Opus 4.8 on trigger.
• A claude-opus-5 string was reportedly spotted on Google Vertex AI Model Garden on July 14, 2026 — unconfirmed by Anthropic or Google.
• Community-guessed release window: late July to early August 2026, possibly before Fable 5 subscription access changes on July 19, 2026.
• Pricing is unconfirmed. Speculation ranges from cheaper than Fable 5's $10/$50 per million tokens to roughly double Opus 4.8's pricing.

What Is Claude Opus 5?

Claude Opus 5 is the anticipated next entry in Anthropic's Opus tier, the "most capable" size class within the Claude model family. Anthropic has shipped four Opus 4.x releases in the current generation — 4.5 in November 2025, then 4.6, 4.7, and 4.8 through mid-2026 — but has not yet moved Opus to a 5.x version number, even as Sonnet crossed to Sonnet 5 on June 30, 2026 and the new Mythos-class Fable 5 launched June 9, 2026.

As of July 15, 2026, "Claude Opus 5" is a name used in community discussion, leak posts, and third-party trackers. Anthropic itself has published no model card, no pricing page entry, and no API identifier under that name. Wikipedia's Claude entry lists "Claude Honeycomb" as a stable release dated July 14, 2026, alongside Sonnet 5, Fable 5, Opus 4.8, and Haiku 4.5, per the current Claude (AI) article) <!-- rel:dofollow class:authoritative --> — but the Wikipedia listing predates any Anthropic confirmation and appears to reflect the same third-party sightings covered in this page.

The status, in one sentence: Claude Opus 5 is an unreleased Anthropic model whose existence is inferred from a Cursor leak, a reported Vertex AI listing, and Anthropic's obvious product gap between Opus 4.8 and Fable 5.

Claude Opus 5 at a Glance

Field	Value
Developer	Anthropic
Type	Large language model, Opus-tier flagship
Modality	Text in, text out (vision expected based on Opus 4.x pattern; unconfirmed)
Codename	"Honeycomb" / "Honeycomb EAP" (unconfirmed as Opus 5)
Context window	1M tokens (per Cursor sighting leaks); 2M claimed in some later posts, unconfirmed
Reasoning modes	Includes an "xhigh" (extra high) effort setting, per leaks
Pricing	Not yet confirmed
API model ID	Not yet published
Release date	Not yet announced; community window: July–early August 2026
Availability	Not yet available; claude-opus-5 reportedly listed on Vertex AI Model Garden (unconfirmed)
License	Proprietary (following prior Claude models)
Safety behavior	Reported per-turn controls with fallback to Claude Opus 4.8 on safety trigger

How Claude Opus 5 Works — What Makes It Different

The clearest technical signal comes from the brief Cursor appearance of Claude Honeycomb EAP on July 9, 2026. According to community screenshots and write-ups, the model's description read "Anthropic research model with per-turn controls and safety fallbacks," with a 1 million token context window and an "extra high" reasoning effort setting. When a safety fallback triggered, the request was routed to Claude Opus 4.8 — a chain that only makes sense if Honeycomb sits above Opus 4.8 in capability.

Three coined feature anchors from the leaks are worth remembering, because subsequent write-ups will cite them:

• Honeycomb EAP — the Cursor-facing codename, where "EAP" reads as an early-access program label.
• xhigh effort mode — a reasoning-effort setting above the existing high tier, hinting at longer deliberation budgets for hard tasks.
• Per-turn controls with Opus 4.8 fallback — an architectural pattern where a stronger model gracefully hands off to a smaller confirmed model when safety classifiers fire.

Beyond those three specifics, everything about training data, parameter counts, or benchmark numbers is unpublished. Any post claiming exact SWE-bench, MMLU, or GPQA scores for Opus 5 today is inferring from vibes.

Source: @LuminaXspace

What You Can Do With Claude Opus 5

Based on the leaked positioning, Claude Opus 5 is being aimed at the same workload Opus 4.8 already serves today, with an emphasis on long-running agents and coding. The LuminaXspace leak summary lists the expected focus as "long-running agents and advanced coding" at a capability closer to Anthropic's Fable/Mythos tier.

Practical use cases developers are pre-positioning for:

• Multi-hour autonomous coding runs inside Claude Code and Cursor, where Opus 4.8 currently caps out on long-horizon reliability.
• Whole-repository refactors and code review that benefit from the reported 1M-token context.
• Agentic tool-use loops (plan → call tool → read result → continue) with the xhigh effort mode reserved for hard planning steps.
• High-stakes reasoning that today gets routed to Fable 5 at $10/$50 per million tokens — Opus 5 is expected to be the cheaper alternative when it lands.

None of this is verified against benchmarks. It reflects where Anthropic's product line has an obvious gap and where the Opus tier has historically won share.

How Claude Opus 5 Compares

The cleanest comparison is inside Anthropic's own current lineup. Confirmed rows come from Anthropic's live pricing and model pages; the Opus 5 row is speculative and clearly marked.

Model	Tier	Release	Input / Output per 1M tokens	Context
Claude Fable 5	Mythos	Jun 9, 2026	$10 / $50	1M
Claude Opus 5	Opus	Not yet released	Not yet confirmed	1M (leaked)
Claude Opus 4.8	Opus	May 28, 2026	$5 / $25	200K–1M (per Anthropic docs)
Claude Sonnet 5	Sonnet	Jun 30, 2026	$3 / $15 (post Aug 31)	1M
Claude Haiku 4.5	Haiku	Oct 15, 2025	$1 / $5	Standard

Bindu Reddy's read on X captures the pricing risk: Claude Opus 5 "is coming as soon as next week... the worst case is it costs 2x that of Opus 4.8". A community post from @ChrissGPT points the other way, claiming Opus 5 will be "much cheaper than Fable 5" while beating GPT-5.6 Sol and Terra on SWE Pro. Both are speculation. See our Claude Sonnet 5 deep dive for how Anthropic has been pricing the Sonnet-to-Opus gap this generation, and our Claude Fable 5 return analysis for context on the Mythos-tier that Opus 5 will sit beneath.

Availability: How to Access Claude Opus 5

Claude Opus 5 is not generally available. There is no public API endpoint, Claude.ai selector option, or claude-opus-5 model ID confirmed by Anthropic as of July 15, 2026.

Two pre-release access signals are worth tracking:

• Cursor model picker — Claude Honeycomb EAP briefly appeared on July 9, 2026 and was removed within hours. Similar transient sightings preceded prior Anthropic releases; they also sometimes precede models that never publicly ship.
• Google Vertex AI Model Garden — a claude-opus-5 entry was reportedly seen on Vertex on July 14, 2026, per a prediction from @kimmonismus that "Claude Opus 5 on vertex means release is kinda imminent." Neither Anthropic nor Google has confirmed the listing.

When Claude Opus 5 does launch, expect it on Anthropic's own API, Claude.ai (Pro, Max, Team, Enterprise tiers), AWS Bedrock, and Google Vertex AI, following the release pattern of Opus 4.5 through 4.8. Compare our earlier reporting on how Anthropic is shipping cross-platform this generation.

What We Don't Know Yet

The reference gaps on this model are large enough that any near-term update to this page will be substantive. Open questions:

• Name. Will it ship as claude-opus-5, claude-opus-4-9, or retain a codename like Honeycomb? Anthropic has not committed to the "5" jump for the Opus tier.
• Exact context window. Leaks split between 1M (majority) and 2M tokens (some later posts). Only Anthropic's model card will settle this.
• Pricing. No confirmed dollar figure exists. Community estimates span roughly $5/$25 to $10/$50 per million tokens.
• Benchmarks. No first-party SWE-bench, GPQA, or MMLU numbers are public. Claims that Opus 5 "beats GPT-5.6 Sol on SWE Pro" originate in an X post from @ChrissGPT without published evidence.
• Modality. Vision, PDF, and tool-use behavior have not been described in any leak.
• Release date. Community targets cluster around late July to early August 2026, but Anthropic has published no calendar.
• Fable 5 relationship. Whether Opus 5 sits below Fable 5, matches it, or eventually replaces the Fable/Mythos naming entirely is unresolved.

Frequently Asked Questions

Is Claude Opus 5 out yet?

Claude Opus 5 is not officially out as of July 15, 2026. Anthropic has not published a model card, an API ID, or an announcement. The current Opus flagship on the API is Claude Opus 4.8, released May 28, 2026.

What is Claude Honeycomb, and is it Claude Opus 5?

Claude Honeycomb is an unreleased Anthropic research model that briefly appeared inside Cursor on or around July 9, 2026 with a 1M-token context window and an "xhigh" reasoning effort mode. Community leaks link Honeycomb to Claude Opus 5, but Anthropic has not confirmed the mapping. It may also ship as Opus 4.9, a Fable variant, or under an entirely different retail name.

When will Claude Opus 5 be released?

Claude Opus 5 has no announced release date. Community leaks point to a July or early August 2026 window, with some posts speculating a launch before Fable 5's subscription access changes on July 19, 2026. None of these dates come from Anthropic and should be treated as unconfirmed.

How much will Claude Opus 5 cost?

Claude Opus 5 pricing has not been announced. Community speculation ranges from "much cheaper than Fable 5" (below $10 per million input tokens) to a worst case of roughly 2x Opus 4.8, which would put it near $10 input and $50 output per million tokens. For reference, Opus 4.8 costs $5 input and $25 output per million tokens.

What is Claude Opus 5's context window?

Leaks around the Honeycomb model report a 1 million token context window, matching Claude Fable 5 and Claude Sonnet 5. Some later community posts claim 2 million tokens. Neither figure has been confirmed by Anthropic.

Claude Opus 5 vs Claude Fable 5 — which is better?

Claude Fable 5 is the confirmed top-tier Mythos-class model today, priced at $10 input and $50 output per million tokens with a 1M context window. Claude Opus 5 is expected to sit in the Opus tier below Fable 5 but above Opus 4.8, closing the gap on agentic coding while costing less than Fable. No head-to-head benchmarks are public.

How will developers access Claude Opus 5?

Once released, Claude Opus 5 is expected on Anthropic's API, Claude.ai, Google Vertex AI, and AWS Bedrock, following the pattern of Opus 4.5 through 4.8. A claude-opus-5 entry was reportedly spotted on Vertex AI Model Garden on July 14, 2026, though this has not been officially confirmed by Anthropic or Google.

What to Watch Next

Three concrete signals will confirm or refute the current picture. First, an Anthropic-hosted claude-opus-5 (or claude-honeycomb) model card and pricing page entry — everything else is downstream of that. Second, whether Fable 5 subscription access on July 19, 2026 gets swapped for an Opus 5 selector, which would validate the July release theory. Third, first-party benchmark numbers versus GPT-5.6 Sol and Terra on SWE-bench and long-horizon agent evals — the only way to test the "Opus 5 catches Fable" claim.

Building similar frontier-agent and long-context coding workloads? On kie.ai you can try Claude Sonnet 5, Claude Fable 5, and Claude Opus 4.8.